WO2022012687A1 - 医疗数据处理方法与系统 - Google Patents

医疗数据处理方法与系统 Download PDF

Info

Publication number
WO2022012687A1
WO2022012687A1 PCT/CN2021/106971 CN2021106971W WO2022012687A1 WO 2022012687 A1 WO2022012687 A1 WO 2022012687A1 CN 2021106971 W CN2021106971 W CN 2021106971W WO 2022012687 A1 WO2022012687 A1 WO 2022012687A1
Authority
WO
WIPO (PCT)
Prior art keywords
text
image
type
data
error
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/106971
Other languages
English (en)
French (fr)
Inventor
高静
董瑞智
姚远波
徐冰
张辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan United Imaging Healthcare Co Ltd
Original Assignee
Wuhan United Imaging Healthcare Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202010690317.8A external-priority patent/CN111950262A/zh
Priority claimed from CN202011434314.4A external-priority patent/CN112530550A/zh
Priority claimed from CN202011495360.5A external-priority patent/CN112527923A/zh
Application filed by Wuhan United Imaging Healthcare Co Ltd filed Critical Wuhan United Imaging Healthcare Co Ltd
Priority to EP21843125.2A priority Critical patent/EP4170670A4/en
Publication of WO2022012687A1 publication Critical patent/WO2022012687A1/zh
Priority to US18/155,710 priority patent/US12562257B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H15/00—ICT specially adapted for medical reports, e.g. generation or transmission thereof
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates

Definitions

  • This specification relates to the medical field, in particular to a medical data processing method and system.
  • One aspect of the present application provides a method implemented on an apparatus including at least one storage device and at least one processing device.
  • the method may include: acquiring a target image; cutting out a first characteristic image from the target image, wherein the first characteristic image includes an image of a region of interest in the target image; acquiring and matching the first characteristic image Image matching second feature image; acquiring a knowledge graph, where the knowledge graph includes the association relationship between the second feature image and the corresponding image description information and diagnosis result information; based on the knowledge graph and the second feature image, find out image description information and diagnosis result information corresponding to the second characteristic image; generating an image report of the target image according to the target image, the image description information and the diagnosis result information.
  • constructing the knowledge graph includes: determining a graph structure of the knowledge graph; acquiring original text and original images in a preset field; and extracting a plurality of entity data from the original text and the original images ; determine the association relationship between the plurality of entity data; construct the domain knowledge graph according to the graph structure, the plurality of entity data and the association relationship between the plurality of entity data.
  • the extracting multiple entity data from the original text and the original image includes: inputting the original text into a pre-trained word segmentation model, and obtaining multiple data output from the word segmentation model. inputting the plurality of word segmentation results into a pre-trained entity recognition model to obtain feature text output by the entity recognition model, and determining the feature text as the entity data.
  • the determining the association relationship between the plurality of entity data includes: inputting the original text and a plurality of the feature texts into a pre-trained relationship extraction model to obtain the relationship extraction model The association relationship between a plurality of the feature texts output by the model.
  • the extracting a plurality of entity data from the original text and the original image includes: inputting the original image into a pre-trained recognition model to obtain the recognition output by the recognition model As a result, the identification result is used to indicate the region of interest in the original image; the characteristic image of the region of interest is cut out from the original image; and the characteristic image of the region of interest is determined as the Describe entity data.
  • the extracting multiple entity data from the original text and the original image, and determining the association relationship between the multiple entity data includes: extracting all the entity data from the original text the image description text and diagnosis result text corresponding to the original image; extract the characteristic image from the original image; determine the image description text, the diagnosis result text, and the characteristic image of the original image as the the entity data; establish an association relationship between the characteristic image of the original image, the image description text and the diagnosis result text.
  • the constructing the knowledge graph according to the graph structure, the plurality of entity data and the association relationship between the plurality of entity data includes: according to the relationship between the plurality of entity data The association relationship forms the multiple entity data into multiple triples; the multiple triples are filled into the graph structure to obtain the knowledge graph.
  • the method may further include: based on the domain knowledge graph and the second feature image, searching for disease information, inducement information, symptom information, drug information or treatment corresponding to the second feature image at least one of the program information; generating an image report of the target image according to at least one of the target image, the disease information, the inducement information, the symptom information, the drug information or the treatment program information.
  • the method may further include: performing error detection on the image report of the target image to obtain the type and location of the erroneous text; A target error correction strategy corresponding to the type of the error text; error correction processing is performed on the error text according to the target error correction strategy and the position of the error text.
  • the type of the erroneous text includes at least one of a typo type, a missing type, and an out-of-order type.
  • Perform error correction processing on the text including: determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text; Error correction processing.
  • determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text including: Extracting the text of the typo type in the image report according to the position of the text of the typo type; selecting from the first database the homophone of the text of the typo type, and/or, the same as the text of the typo type from the first database The text whose similarity between the texts is greater than the similarity threshold is used as the candidate text.
  • determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text including: According to the position of the text of the missing type, the text adjacent to the text of the missing type is extracted from the image report; the text matching the adjacent text is selected from the second database as the candidate text.
  • determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text including : extract the text of the disordered type from the image report according to the position of the text of the disordered type; select the text containing the same word as the text of the disordered type from the third database as the text of the disordered type candidate text.
  • performing error correction processing on the erroneous text according to the at least one candidate text includes: sequentially adding at least one of the candidate texts to positions corresponding to the erroneous text to obtain the text to be processed ; score the smoothness of the text to be processed; sort a plurality of scored texts to be processed according to the order of the scores from high to low, and determine the text to be processed with the highest score as the corrected text.
  • the type of the erroneous text includes a redundant type
  • performing error correction processing on the erroneous text according to the target error correction strategy and the position of the erroneous text includes: according to the The target error correction strategy and the position of the erroneous text determine whether to delete the redundant type of text; if it is determined to delete, the erroneous text is deleted; if it is determined not to delete, the deletion operation is not performed.
  • the determining whether to delete the redundant type of text according to the target error correction strategy and the position of the erroneous text includes: according to the position of the redundant type of text in the to-be-to-be-typed text Extract the text containing the redundant type from the image report; check whether the text containing the redundant type is smooth; if it is not smooth, determine to delete the text of the redundant type; if it is smooth, determine not to delete the text Redundant type of text.
  • the image report is unstructured medical data
  • the method further includes: parsing the image report to obtain structured medical data corresponding to the image report, and the image report and the image report.
  • the corresponding relationship between the structured medical data is stored; the structured medical data and the corresponding relationship are stored in a fourth database, and the image report is stored in a fifth database.
  • the analyzing the image report to obtain structured medical data corresponding to the image report, and the corresponding relationship between the image report and the structured medical data include: According to the unstructured medical data type of the image report, the corresponding parsing rules are searched from a plurality of preset parsing rules; according to the corresponding parsing rules, the image report is parsed, and the structured image report corresponding to the image report is obtained. Medical data; establishing a correspondence between the image report and the structured medical data according to the identification of the unstructured medical data in the image report and the identification of the structured medical data.
  • the plurality of preset parsing rules include: medical image text recognition rules, DICOM file parsing rules and XML report parsing rules.
  • the method further includes: receiving the image report sent by the corresponding medical device through a preset application program interface; or, according to the target address identifier, in the medical device corresponding to the target address identifier A search is performed to obtain the image report.
  • the method further includes: receiving a query request, where the query request carries the data identifier of the image report; based on the query request, according to the data identifier and the corresponding relationship, The fourth database and the fifth database are queried to obtain the structured medical data and the image report corresponding to the data identifier.
  • the fourth database and the fifth database are respectively queried to obtain the structure corresponding to the data identifier.
  • the medical data and the image report include: based on the query request, performing a query in the fourth database according to the data identifier, and obtaining structured medical data corresponding to the data identifier; The structured medical data and the corresponding relationship are queried in the fifth database to obtain the image report.
  • the at least one storage device includes a set of instructions.
  • the at least one processor is in communication with the at least one storage device, wherein when the at least one processor executes the set of instructions, the at least one processor is configured to cause the system to perform at least one operation, including: obtaining a target image; cutting out a first characteristic image from the target image, wherein the first characteristic image includes an image of an area of interest in the target image; acquiring a second characteristic image matching the first characteristic image Obtaining a knowledge map, the knowledge map includes the association relationship between the second feature image and the corresponding image description information and diagnosis result information; based on the knowledge map and the second feature image, find out the relationship between the second feature image and the second feature image Corresponding image description information and diagnosis result information; an image report of the target image is generated according to the target image, the image description information and the diagnosis result information.
  • Another aspect of the present application provides a system, comprising: an image capturing module configured to acquire a target image; and capturing a first characteristic image from the target image, wherein the first characteristic image includes the target An image of the region of interest in the image.
  • the system may further include an image search module configured to obtain a second characteristic image that matches the first characteristic image.
  • the system may further include an information search module configured to acquire a knowledge graph, the knowledge graph including the association relationship between the second feature image and the corresponding image description information and diagnosis result information; and based on the knowledge graph and the first
  • the second characteristic image is to find out the image description information and diagnosis result information corresponding to the second characteristic image.
  • the system may further include a report generation module configured to generate an image report of the target image based on the target image, the image description information and the diagnosis result information.
  • Another aspect of the present application provides a non-transitory computer-readable medium comprising at least one set of instructions, wherein the at least one set of instructions when executed by one or more processors of a computing device, the at least one set of instructions
  • the set causes the computing device to perform a method, the method comprising: acquiring a target image; and extracting a first feature image from the target image, wherein the first feature image includes a region of interest in the target image obtain a second feature image that matches the first feature image; obtain a knowledge map, where the knowledge map includes the association relationship between the second feature image and the corresponding image description information and diagnosis result information; based on the knowledge map and the second characteristic image, to find out the image description information and diagnosis result information corresponding to the second characteristic image; generate the target image according to the target image, the image description information and the diagnosis result information Video report.
  • Another aspect of the present application provides a data processing method implemented on an apparatus including at least one storage device and at least one processing device.
  • the method includes: performing error detection on the text to be processed to obtain the type and position of the error text; determining a target error correction strategy corresponding to the type of the error text according to the correspondence between the error text type and the error correction strategy; The target error correction strategy and the position of the erroneous text are used to perform error correction processing on the erroneous text.
  • the at least one storage device includes a set of instructions.
  • the at least one processor is in communication with the at least one storage device, wherein when the at least one processor executes the set of instructions, the at least one processor is configured to cause the system to perform at least one operation, including: treating Process the text to perform error detection to obtain the type and position of the wrong text; according to the correspondence between the wrong text type and the error correction strategy, determine the target error correction strategy corresponding to the type of the wrong text; According to the target error correction strategy and the position of the erroneous text, and perform error correction processing on the erroneous text.
  • Another aspect of the present application provides a system, comprising: a detection module, configured to perform error detection on the text to be processed, to obtain the type and location of the error text; a determination module, configured to determine the difference between the type of the error text and the error correction strategy The corresponding relationship, determine the target error correction strategy corresponding to the type of the error text; the error correction module is configured to perform error correction processing on the error text according to the target error correction strategy and the position of the error text .
  • Another aspect of the present application provides a non-transitory computer-readable medium comprising at least one set of instructions, wherein the at least one set of instructions when executed by one or more processors of a computing device, the at least one set of instructions
  • the set causes the computing device to execute a method, the method comprising: performing error detection on the text to be processed to obtain the type and location of the erroneous text; The target error correction strategy corresponding to the type of text; according to the target error correction strategy and the position of the error text, error correction processing is performed on the error text.
  • Another aspect of the present application provides an unstructured data storage method, implemented on a device including at least one storage device and at least one processing device, the method comprising: acquiring unstructured medical data to be stored; Parse the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data, and the corresponding relationship between the unstructured medical data and the structured medical data; The data and the corresponding relationship are stored in a fourth database, and the unstructured medical data is stored in a fifth database.
  • the at least one storage device includes a set of instructions.
  • the at least one processor is in communication with the at least one storage device, wherein when the at least one processor executes the set of instructions, the at least one processor is configured to cause the system to perform at least one operation, including: obtaining Unstructured medical data to be stored; analyze the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data, as well as the unstructured medical data and the structured medical data The corresponding relationship between the structured medical data and the corresponding relationship is stored in the fourth database, and the unstructured medical data is stored in the fifth database.
  • Another aspect of the present application provides a system, comprising: an acquisition module configured to acquire unstructured medical data to be stored; a parsing module configured to parse the unstructured medical data to obtain the unstructured medical data Structured medical data corresponding to structured medical data, and the corresponding relationship between the unstructured medical data and the structured medical data; a storage module configured to store the structured medical data and the corresponding relationship
  • the unstructured medical data is stored in the fourth database, and the unstructured medical data is stored in the fifth database.
  • Another aspect of the present application provides a non-transitory computer-readable medium comprising at least one set of instructions, wherein the at least one set of instructions when executed by one or more processors of a computing device, the at least one set of instructions set to make the computing device execute a method, the method includes: acquiring unstructured medical data to be stored; parsing the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data data, and the corresponding relationship between the unstructured medical data and the structured medical data; store the structured medical data and the corresponding relationship in a fourth database, and store the unstructured medical data stored in the fifth database.
  • FIG. 1 is a schematic diagram of an application scenario of a medical system according to some embodiments of the present application.
  • FIG. 2 is a schematic diagram of exemplary hardware and/or software of a computing device according to some embodiments of the present application
  • FIG. 3 is a schematic diagram of exemplary hardware and/or software of a mobile device according to some embodiments of the present application.
  • FIG. 4 is an exemplary flowchart of a method for generating an image report according to some embodiments of the present application
  • FIG. 5 is an exemplary flowchart of steps of constructing a domain knowledge graph according to some embodiments of the present application.
  • FIG. 6 is a schematic diagram of a map structure according to some embodiments of the present application.
  • FIG. 7 is a schematic diagram of a domain knowledge graph according to some embodiments of the present application.
  • FIG. 8 is an exemplary flowchart of searching for image description information and diagnosis result information corresponding to a target image according to some embodiments of the present application
  • FIG. 9 is an exemplary flowchart of generating an image report according to some embodiments of the present application.
  • FIG. 10 is a schematic diagram of an image report according to some embodiments of the present application.
  • FIG. 11 is an exemplary block diagram of an image report generating apparatus according to some embodiments of the present application.
  • 12A is an exemplary flowchart of a text error correction method according to some embodiments of the present application.
  • FIG. 12B is an exemplary flowchart of a training method according to some embodiments of the present application.
  • 13 is an exemplary flowchart of performing error correction processing on error text according to a target error correction strategy and the position of the error text according to some embodiments of the present application;
  • FIG. 14 is an exemplary flowchart of determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text according to some embodiments of the present application;
  • 15 is an exemplary flowchart of determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text according to some embodiments of the present application;
  • 16 is an exemplary flowchart of determining at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text according to some embodiments of the present application;
  • 17 is an exemplary flowchart of performing error correction on erroneous text according to a target error correction strategy and the position of erroneous text according to some embodiments of the present application;
  • 18 is an exemplary flowchart of performing error correction processing on erroneous text according to a target error correction strategy and the position of erroneous text according to some embodiments of the present application;
  • FIG. 20 is an exemplary flowchart of a data processing method according to some embodiments of the present application.
  • 21 is a schematic diagram of the structure of an error correction system according to some embodiments of the present application.
  • Fig. 22 is a block diagram showing the structure of a data processing apparatus according to some embodiments of the present application.
  • FIG. 23 is an exemplary flowchart of a method for storing unstructured data according to some embodiments of the present application.
  • FIG. 24 is an exemplary flowchart of a method for storing unstructured data according to some embodiments of the present application.
  • FIG. 25 is an exemplary flowchart of a method for storing unstructured data according to some embodiments of the present application.
  • FIG. 26A is an exemplary flowchart of a method for storing unstructured data according to some embodiments of the present application.
  • FIG. 26B is an exemplary flowchart of a method for storing unstructured data according to some embodiments of the present application.
  • FIG. 27 is a structural block diagram of an unstructured data storage device according to some embodiments of the present application.
  • Figure 28 is an exemplary flowchart of a medical system workflow according to some embodiments of the present application.
  • the present application provides a method, device, computer equipment and storage medium for generating an image report, acquiring a target image, and cutting out a first characteristic image from the target image; searching in a preset characteristic image library according to the first characteristic image , obtain a second feature image matching the first feature image; based on the pre-built domain knowledge map and the second feature image, find out the image description information and diagnosis result information corresponding to the target image; according to the target image, image description information and The diagnostic result information generates an image report.
  • the terminal can automatically generate an image report by using a pre-built domain knowledge graph. Compared with the prior art, it is not necessary for a doctor to manually fill in the image report, which improves the generation efficiency of the image report.
  • the present application provides a data processing method, device, computer equipment and storage medium.
  • the type and position of the erroneous text are obtained, and according to the corresponding relationship between the erroneous text type and the error correction strategy, Determine the target error correction strategy corresponding to the type of the error text, and further perform error correction processing on the error text according to the target error correction strategy and the position of the error text.
  • the error correction data processing method since each type of error text corresponds to its own target error correction strategy, the error correction operation for each type of error text is realized, and compared with the traditional error text regardless of any type, the error correction operation is realized.
  • the use of a large database to obtain texts for error correction leads to the problems of large data processing volume and poor error correction accuracy.
  • the error correction method described in this application greatly improves the error correction accuracy and data processing speed.
  • Another aspect of the present application provides an unstructured data storage method, device, computer equipment and storage medium.
  • structured medical data corresponding to the unstructured medical data can be obtained, and The corresponding relationship between the unstructured medical data and the obtained structured medical data, so that the obtained structured medical data and the obtained corresponding relationship can be stored in the preset first database, and the unstructured medical data to be stored can be stored.
  • the chemical medical data is stored in the preset second database. Since the process of parsing the unstructured medical data is very simple, the structured medical data corresponding to the unstructured medical data to be stored can be quickly and accurately obtained, and the unstructured medical data corresponding to the unstructured medical data can be obtained quickly and accurately.
  • the corresponding relationship between the structured medical data and the obtained structured medical data so that the obtained structured medical data and the obtained corresponding relationship can be quickly stored in the preset first database, and the unstructured medical data to be stored can be stored
  • the medical data is stored in the preset second database, thereby improving the storage efficiency of the unstructured medical data to be stored.
  • FIG. 1 is a schematic diagram of an application scenario of a medical system 100 according to some embodiments of the present application.
  • medical system 100 may include data processing system 110 , network 120 , user terminal 130 , and storage device 140 .
  • Data processing system 110 may be used to process medical data for medical system 100 .
  • Data processing system 110 may include image archiving and transmission system (PACS), post-processing workstation, surgical system, hospital information system (HIS), laboratory information system (LIS), radiology information system (RIS), doctor workstation, nurse workstation , One or a combination of imaging systems, hospitalization management systems, outpatient and emergency management systems, and charging systems.
  • PPS image archiving and transmission system
  • HIS hospital information system
  • LIS laboratory information system
  • RIS radiology information system
  • doctor workstation nurse workstation
  • One or a combination of imaging systems hospitalization management systems, outpatient and emergency management systems, and charging systems.
  • data processing system 110 may include one or more servers.
  • the server may be used to manage resources and process data and/or information from at least one component of the system or external systems.
  • the servers may be stand-alone servers or groups of servers.
  • the server group may be centralized or distributed (eg, the server may be a distributed system).
  • the server may be local or remote in some embodiments.
  • a server may access information and/or materials stored in storage device 140 via network 120 .
  • the server may execute on a cloud platform.
  • the cloud platform may include one or any combination of a private cloud, a public cloud, a hybrid cloud, a community cloud, a decentralized cloud, an internal cloud, and the like.
  • the server may include a processing device.
  • a processing device may include one or more sub-processing devices (eg, a single-core processing device or a multi-core multi-core processing device).
  • a processing device may include a central processing unit (CPU), an application specific integrated circuit (ASIC), an application specific instruction processor (ASIP), a graphics processing unit (GPU), a physical processor (PPU), a digital signal processor (DSP) ), field programmable gate array (FPGA), programmable logic circuit (PLD), controller, microcontroller unit, reduced instruction set computer (RISC), microprocessor, etc. or any combination of the above.
  • CPU central processing unit
  • ASIC application specific integrated circuit
  • ASIP application specific instruction processor
  • GPU graphics processing unit
  • PPU physical processor
  • DSP digital signal processor
  • FPGA field programmable gate array
  • PLD programmable logic circuit
  • controller microcontroller unit, reduced instruction set computer (RISC), microprocessor, etc. or any combination of the above.
  • Data and/or information may be exchanged between components of the medical system 100 through the network 120 .
  • one or more components in medical system 100 eg, data processing system 110 , user terminal 130 , storage device 140
  • network 120 may be any type of wired or wireless network.
  • the network 120 may include a cable network, a wired network, a fiber optic network, a telecommunications network, an internal network, the Internet, a local area network (LAN), a wide area network (WAN), a wireless area network (WLAN), a metropolitan area network (MAN) , Public Switched Telephone Network (PSTN), Bluetooth network, ZigBee network, Near Field Communication (NFC) network, etc. or any combination of the above.
  • network 120 may include one or more network entry and exit points.
  • network 120 may include wired or wireless network entry and exit points, such as base stations and/or Internet exchange points, through which one or more components of medical system 100 may connect to network 120 to exchange data and/or information .
  • User terminal 130 may perform data interaction with data processing system 110 .
  • user terminal 130 may receive data and/or information from storage device 140 and/or data processing system 110 .
  • the user terminal 130 may include a mobile phone (141-1, 141-2, 141-3%), a tablet computer (142-1, 142-2, 142-3%), a laptop computer (143-1, 143-2, 143-3%) etc. or any combination thereof.
  • user terminal 130 may establish communication with data processing system 110 through application software in user terminal 130 .
  • user terminal 130 may receive, display data and/or information from storage device 140 and/or data processing system 110 through application software.
  • the user terminal 130 may send instructions and/or requests to the storage device 140 and/or the data processing system 110 through application software.
  • One or more components of medical system 100 may access data or instructions stored in storage device 140 via network 120 .
  • Data refers to the digital representation of information, and can include various types, such as binary data, text data, image data, video data, and so on. Instructions are programs that control a device or device to perform a specific function.
  • Storage device 140 may be connected to network 120 to communicate with one or more components of medical system 100 (eg, data processing system 110, user terminal 130). In some embodiments, storage device 140 may be part of a server. Storage device 140 may be implemented in a single central server, multiple servers connected by communication links, or multiple personal devices. The storage device 140 may be generated by a number of personal devices and cloud servers.
  • storage device 140 may include mass memory, removable memory, volatile read-write memory (eg, random access memory RAM), read only memory (ROM), the like, or any combination thereof.
  • the database may be implemented on a cloud platform.
  • the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a decentralized cloud, an internal cloud, etc., or any combination thereof.
  • computing device 200 may include processor 210 , memory 220 , input/output (I/O) interface 230 , and communication port 240 .
  • processor 210 may include processor 210 , memory 220 , input/output (I/O) interface 230 , and communication port 240 .
  • I/O input/output
  • communication port 240 may be implemented in accordance with processing device 200 .
  • the processor 210 may execute computational instructions (program code) and perform the functions of the medical system 100 described herein.
  • the computing instructions may include programs, objects, components, data structures, procedures, modules, and functions (the functions refer to the specific functions described in this application).
  • processor 210 may process target images obtained from any component of medical system 100 .
  • processor 210 may include a microcontroller, microprocessor, reduced instruction set computer (RISC), application specific integrated circuit (ASIC), application specific instruction set processor (ASIP), central processing unit (CPU) , Graphics Processing Units (GPUs), Physical Processing Units (PPUs), Microcontroller Units, Digital Signal Processors (DSPs), Field Programmable Gate Arrays (FPGAs), Advanced RISC Machines (ARMs), Programmable Logic Devices, and Any circuits, processors, etc., or any combination thereof, that perform one or more functions.
  • the computing device 200 in FIG. 2 only depicts one processor, but it should be noted that the computing device 200 in this application may also include multiple processors.
  • Memory 220 may store data/information obtained from any other component of medical system 100 .
  • memory 220 may include mass storage, removable storage, volatile read and write memory, read only memory (ROM), etc., or any combination thereof.
  • Exemplary mass storage may include magnetic disks, optical disks, solid state drives, and the like.
  • Removable storage may include flash drives, floppy disks, optical disks, memory cards, compact disks, magnetic tapes, and the like.
  • Volatile read and write memory may include random access memory (RAM).
  • RAM can include Dynamic RAM (DRAM), Double Rate Synchronous Dynamic RAM (DDR SDRAM), Static RAM (SRAM), Thyristor RAM (T-RAM), and Zero Capacitance (Z-RAM), among others.
  • DRAM Dynamic RAM
  • DDR SDRAM Double Rate Synchronous Dynamic RAM
  • SRAM Static RAM
  • T-RAM Thyristor RAM
  • Z-RAM Zero Capacitance
  • ROM may include mask ROM (MROM), programmable ROM (PROM), erasable programmable ROM (PEROM), electrically erasable programmable ROM (EEPROM), compact disk ROM (CD-ROM), and digital versatile disk ROM Wait.
  • MROM mask ROM
  • PROM programmable ROM
  • PEROM erasable programmable ROM
  • EEPROM electrically erasable programmable ROM
  • CD-ROM compact disk ROM
  • digital versatile disk ROM Wait digital versatile disk ROM Wait.
  • An input/output interface (I/O) 230 may be used to input or output signals, data or information. In some embodiments, input/output interface 230 may allow a user to communicate with medical system 100 . In some embodiments, input/output interface (I/O) 230 may include input devices and output devices. Exemplary input devices may include one or any combination of keyboards, mice, touch screens, microphones, and the like. Exemplary output devices may include display devices, speakers, printers, projectors, etc., or any combination thereof. Exemplary display devices may include one or any combination of liquid crystal displays (LCDs), light emitting diode (LED) based displays, flat panel displays, curved displays, television equipment, cathode ray tubes (CRTs), and the like.
  • LCDs liquid crystal displays
  • LED light emitting diode
  • CRTs cathode ray tubes
  • Communication port 240 may connect to a network for data communication.
  • the connection may be a wired connection, a wireless connection, or a combination of both.
  • Wired connections may include electrical cables, fiber optic cables, or telephone lines, etc., or any combination thereof.
  • Wireless connections may include one or any combination of Bluetooth, Wi-Fi, WiMax, WLAN, ZigBee, mobile networks (eg, 3G, 4G, or 5G, etc.), and the like.
  • the communication port 240 may be a standardized port such as RS232, RS485, or the like.
  • communication port 240 may be a specially designed port.
  • the communication port 240 may be designed according to the Digital Imaging and Medical Communication Protocol (DICOM).
  • DICOM Digital Imaging and Medical Communication Protocol
  • FIG. 3 is a schematic structural diagram of a mobile device according to some embodiments of the present application, the mobile device can be configured to implement a specific system disclosed in the present application, for example, a user terminal 130.
  • the user equipment for displaying and interacting with location-related information is a mobile device 300, which may include, but is not limited to, smart phones, tablet computers, music players, portable game consoles, Global Positioning System (Global Positioning System, GPS) receivers, wearable computing devices (such as glasses, watches, etc.), or other forms.
  • a mobile device 300 may include, but is not limited to, smart phones, tablet computers, music players, portable game consoles, Global Positioning System (Global Positioning System, GPS) receivers, wearable computing devices (such as glasses, watches, etc.), or other forms.
  • GPS Global Positioning System
  • wearable computing devices such as glasses, watches, etc.
  • the mobile device 300 in this example may include one or more central processing units (CPUs) 340, Graphical Processing Units (GPUs) 330, a display unit 320, a memory 360, an antenna 310 (eg, a wireless communication unit), Storage unit 370, and one or more input/output (I/O) units 350.
  • CPUs central processing units
  • GPUs Graphical Processing Units
  • Any other suitable components which may include, but are not limited to, a system bus or a controller (not shown), may also be included in mobile device 300 .
  • a mobile operating system 361 such as iOS, Android, Windows Phone, etc.
  • one or more applications 362 can be loaded into the memory 360 from the storage unit 370 and executed by the central processing unit 340 .
  • Application 362 may include a browser or other mobile application suitable for receiving and processing image reports on mobile device 300 .
  • User (e.g., physician) interaction with respect to image reports may be obtained through input/output (I/O) unit 350 and provided to server 120, and/or other components of medical system 100, e.g., through network 120.
  • I/O input/output
  • a computer hardware platform may be used as the hardware platform for one or more of the elements described above (eg, a server, and/or a hardware platform of the medical system 100 ). other components).
  • the hardware elements, operating systems, and programming languages of such computers are common in nature, and it is assumed that those skilled in the art are sufficiently familiar with these techniques to be able to use the techniques described herein to provide the information needed for on-demand services.
  • a computer that contains user interface elements can be used as a personal computer (PC) or other type of workstation or terminal device, or as a server when properly programmed.
  • PC personal computer
  • Such structures, procedures, and general operation of such computer equipment are believed to be familiar to those skilled in the art, and therefore no additional explanation is required for any of the figures.
  • the existing image report generation method is to fill the obtained medical images into the template of the image report provided in advance, and the unfilled part in the template is filled in manually by the doctor who read the images.
  • This way of generating image reports is inefficient.
  • the present application provides a method and system for automatically generating an image report.
  • FIG. 4 is an exemplary flowchart of a method for generating an image report according to some embodiments of the present application.
  • process 400 may be performed by medical system 100 (eg, data processing system 110).
  • process 400 may be performed by a PACS, an imaging system, or a server of a post-processing workstation.
  • the process 400 may be performed by the processor 210 .
  • the process 400 may be performed by one or more of the modules shown in FIG. 11 .
  • process 400 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 400 may be implemented.
  • a storage device eg, storage device 140
  • process 400 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order in which the operations of flow 400 are shown in FIG. 4 and described below is not intended to be limiting.
  • the data processing system 110 may acquire the target image, and capture the first characteristic image from the target image.
  • the target image may include a medical image generated by an imaging system, for example, CT (Computed Tomography) image, MR (Magnetic Resonance, magnetic resonance) image, PET (Positron Emission Computed Tomography, positron emission tomography) image At least one of emission computed tomography) images, X-ray films (X-Ray), ultrasound images, and the like.
  • CT Computer Planar Tomography
  • MR Magnetic Resonance, magnetic resonance
  • PET Positron Emission Computed Tomography
  • positron emission tomography At least one of emission computed tomography
  • X-ray films X-Ray
  • ultrasound images and the like.
  • the feature image may refer to the image of the region of interest.
  • the region of interest can be the whole or a part of the detection object (eg, human body, animal, object, etc.), such as head, chest, abdomen, heart, liver, upper limbs, lower limbs, spine, bones, blood vessels, lesions site, tumor site, etc., or any combination of the foregoing sites.
  • the first feature image may be a part of the target image.
  • the target image may be a cross-sectional CT image of the abdomen of the detection object
  • the first characteristic image may be an image of the liver in the abdominal CT image.
  • the target image there may be various ways to acquire the target image, such as acquiring the target image from an imaging device through the network 120, or acquiring the target image from an image database (eg, the storage device 140, the memory 220, PACS, etc.) through the network 120 .
  • the embodiment of the present application does not limit the acquisition method of the target image.
  • the region of interest in the target image can be identified, and then the image in the region of interest is cut out as the first feature image.
  • a method of determining a region of interest may include: inputting a target image into a pre-trained recognition model, and obtaining the region of interest in the target image output by the recognition model.
  • Another way of determining the region of interest may include: the user manually selects the region of interest in the target image on the user terminal 130, the data processing system 110 receives the user's selection operation on the target image from the user terminal 130, and determines according to the selection operation the region of interest in the target image, thereby acquiring the first feature image.
  • the embodiments of the present application do not limit the manner of determining the region of interest.
  • the data processing system 110 may search a preset feature image library according to the first feature image to obtain a second feature image matching the first feature image.
  • the feature image library may include multiple feature images, such as CT (Computed Tomography) images, MR (Magnetic Resonance, magnetic resonance imaging) images. Resonance) image, PET (Positron Emission Computed Tomography, Positron Emission Computed Tomography) image, at least one of X-ray (X-Ray), ultrasound image and the like.
  • CT Computer Tomography
  • MR Magnetic Resonance, magnetic resonance imaging
  • Resonance PET
  • PET Positron Emission Computed Tomography
  • Positron Emission Computed Tomography Positron Emission Computed Tomography
  • ultrasound image at least one of X-ray (X-Ray), ultrasound image and the like.
  • a feature image library is preset, and after the first feature image is cut out from the target image, the data processing system 110 may perform similarity calculation between the first feature image and each feature image in the feature image library, respectively, Get multiple similarities.
  • the data processing system 110 may determine a feature image whose similarity satisfies a preset condition in the feature image library as the second feature image.
  • the preset condition may include that the similarity is greater than a preset threshold (eg, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, etc.).
  • the similarity 1 (eg, 96%) between the first feature image A and the feature image 1 in the feature image library is calculated, and the similarity between the first feature image A and the feature image 2 in the feature image library is calculated 2 (eg, 70%), and so on to calculate multiple similarities.
  • the preset threshold is set to 95%
  • the second feature image for example, the feature image 1
  • the type of region of interest eg, head, chest, abdomen, heart, liver, upper limbs, etc.
  • lower limbs, spine, bones, blood vessels, etc. e.g., morphological features of the region of interest
  • pixel features e.g, gray value
  • the similarity between the feature image in the feature image library and the first feature image may be determined by a similarity algorithm, and an exemplary similarity algorithm may include a mean hash algorithm, a histogram algorithm, and the like.
  • the similarity between the feature image in the feature image library and the first feature image may be determined by inputting the feature image in the feature image library and the first feature image into the trained similarity determination model .
  • the data processing system 110 may search for the image description corresponding to the second characteristic image based on the pre-built domain knowledge graph (also referred to as "knowledge graph”) and the second feature image information and diagnostic result information.
  • the pre-built domain knowledge graph also referred to as "knowledge graph”
  • the image description information may be information used to describe the features of the region of interest.
  • the image description information includes at least one of the shape, size, and texture of the region of interest in the image.
  • the diagnosis result information may be disease diagnosis-related information of the region of interest.
  • the diagnostic result information may include the diagnostic result of the region of interest in the image.
  • the data processing system 110 may further search for at least one of diseases, causes, symptoms, drugs, treatment plans, etc. corresponding to the second characteristic image based on the pre-built domain knowledge graph and the second characteristic image. text.
  • the domain knowledge graph may include correspondence between images and text.
  • the domain knowledge graph may include correspondence between image entities, correspondence between text entities, and/or correspondence between image entities and text entities.
  • Image entities may include original images and/or feature images.
  • the textual entities may include at least one of image descriptions, diagnoses, diseases, causes, symptoms, drugs, treatment regimens, and the like.
  • the domain knowledge graph may include the correspondence between the historical image report and the images in the historical image report.
  • a domain knowledge graph can be pre-built, and after finding a second feature image similar to the first feature image from the feature image library, the data processing system 110 can perform the process in the domain knowledge graph according to the second feature image. Search, find out the image description information and diagnosis result information associated with the second characteristic image, and use these image description information and diagnosis result information as the image description information and diagnosis result information corresponding to the target image.
  • the knowledge graph please refer to Figure 5- Figure 7.
  • the data processing system 110 may generate an image report according to the target image, the image description information and the diagnosis result information.
  • the image report may include basic information of the patient (eg, name, age, gender, examination items, etc.), target image, image description information, diagnosis result information, and the like.
  • an image report template may be preset, and after finding the image description information and diagnosis result information of the target image, the data processing system 110 may report the basic information, target image, image description information and diagnosis result information of the patient to be treated Populate the image report template, and then automatically generate the image report.
  • the imaging report may also include at least one of a disease, a cause, a symptom, a drug, a treatment regimen, and the like.
  • the data processing system 110 can fill at least one of diseases, causes, symptoms, drugs, treatment plans, etc. into the image report template, and then automatically generate an image report.
  • the image report template may be stored in a storage device (eg, storage device 140 , memory 220 , etc.) of the medical system 100 .
  • the data processing system 110 may access the storage device stored in the medical system 100 to obtain the image report template.
  • error detection and error correction may be performed on the generated image report of the target image according to the error correction method of FIGS. 12-22 .
  • the generated image report of the target image is unstructured data, and the generated image report of the target image can be stored according to the unstructured data storage method of FIGS. 23-27 .
  • the data processing system 110 obtains the target image, and cuts out the first characteristic image from the target image; searches in the preset characteristic image database according to the first characteristic image, and obtains a match with the first characteristic image. based on the pre-built domain knowledge map and the second feature image, find out the image description information and diagnosis result information corresponding to the target image; generate an image report according to the target image, image description information and diagnosis result information.
  • the data processing system 110 can automatically generate an image report by using a pre-built domain knowledge graph, without requiring a doctor to manually fill in the image report, which improves the generation efficiency of the image report.
  • process 400 may be performed by user terminal 130, eg, by CPU 340 and/or GPU 330, and/or by one or more of the modules shown in FIG. 11 .
  • FIG. 5 is an exemplary flowchart of steps of constructing a domain knowledge graph according to some embodiments of the present application.
  • process 500 may be performed by medical system 100 (eg, data processing system 110).
  • process 500 may be performed by a PACS, a post-processing workstation, a physician's workstation, or a server of an imaging system.
  • the process 500 may be performed by the processor 210 .
  • process 500 may be performed by one or more of the modules shown in FIG. 11 .
  • process 500 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 500 may be implemented.
  • a storage device eg, storage device 140
  • process 500 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order in which the operations of flow 500 are illustrated in Figure 5 and described below is not intended to be limiting. In some embodiments, a domain knowledge graph may be constructed based on process 500 .
  • the data processing system 110 may determine the graph structure of the domain knowledge graph.
  • the graph structure can be used to characterize entity types and entity connection relationships.
  • the entity types may include image entities and/or text entities.
  • Image entities may include original images and/or feature images.
  • the textual entities may include at least one of image descriptions, diagnoses, diseases, causes, symptoms, drugs, treatment regimens, and the like.
  • the entity connection relationship may represent a relationship between entity types, including at least one of observe, inhibit, juxtapose, select, use, feature, result, relate, complication, contain, induce, manifest, and the like.
  • entity type Attributes original image ID, Build Date, Path Feature image ID, lesion, path diagnostic result ID, name, description image description ID, name, description disease ID, Chinese medicine name, Western medicine name, description, susceptible population, infectious disease Incentive ID, name, description, type symptom ID, Name, Description, Severity, Body Part drug ID, Chinese medicine name, Western medicine name, description treatment plan ID, description, treatment cycle
  • the graph structure shown in FIG. 6 may be determined according to the entity type and entity connection relationship described above.
  • entity types can include image entities and text entities.
  • Image entities can include original images and feature images.
  • Text entities can include image descriptions, diagnoses, diseases, causes, symptoms, medications, treatment options.
  • Entity connection relationships can include observation, restraint, juxtaposition, selection, use, characteristic, result, relation, complication, containment, inducement, manifestation.
  • feature relationship between the original image and the feature image.
  • observation relationship between the characteristic image and the image description.
  • outputcome relationship between characteristic images and diagnostic results.
  • characteristic images and disease There is an “observation” relationship between image descriptions and disease.
  • the data processing system 110 can obtain the original text and original image of the preset field, extract multiple entity data from the original text and original image, and determine the relationship between the multiple entity data. connection relation.
  • the preset field may include the medical field.
  • the manner in which the data processing system 110 obtains the original text and original images may include: obtaining the original text of medical common sense from books, newspapers, and websites; Obtain original texts and original images related to diagnosis from medical records, laboratory tests, and diagnostic reports; obtain original images from medical imaging equipment. This embodiment of the present application does not limit the acquisition method.
  • the data processing system 110 may store the acquired raw text in a MySQL data table, and save the acquired raw image in a DICOM format to an image database (eg, PACS, storage device 140, storage 220, etc.) , and store the save path to the MySQL data table.
  • an image database eg, PACS, storage device 140, storage 220, etc.
  • the acquired original text may be subjected to error correction processing before being stored.
  • the acquired original text can be used as the text to be processed, and the acquired original text can be corrected for errors according to the text error correction methods shown in FIGS. 12 to 22 .
  • the original text and original images obtained from books, newspapers, websites, electronic medical records, laboratory tests, diagnostic reports, etc. may be unstructured data, and may be stored according to the unstructured data in FIGS. 23-27 .
  • the method stores the original text and original image.
  • the data processing system 110 can extract multiple text entity data from the original text, and determine the association relationship between the extracted multiple text entity data; or
  • the image entity data is extracted from the original image, and the relationship between the original image and the image entity data is determined; the text entity data and image entity data can also be extracted from the original text and the original image, and the text entity data and image entity data can be determined.
  • the relationship between entity data can be determined.
  • the data processing system 110 may construct a domain knowledge graph according to the graph structure, the multiple entity data and the association relationship between the multiple entity data.
  • the data processing system 110 may group the plurality of entity data into a plurality of triplet; then, fill multiple triples into the graph structure to obtain a domain knowledge graph, for example, store multiple triples in a preset graph database to obtain a domain knowledge graph.
  • the association between text entity data can be expressed as ⁇ XX pneumonia (disease), use, methylprednisolone (drug)>
  • the text entity The relationship between the data and the image entity data is represented as ⁇ XX pneumonia (disease), observation, 6ff55da4-5454-4b81-a831-a25c7bf68f4d (feature image)>
  • the relationship between the image entity data bracket is represented as ⁇ d3e616c8-1d39 -4fd6-a2fe-e6a7e201732d (original image), feature, 6ff55da4-5454-4b81-a831-a25c7bf68f4d (feature image) >.
  • d3e616c8-1d39-4fd6-a2fe-e6a7e201732d is the unique identifier of an original image
  • 6ff55da4-5454-4b81-a831-a25c7bf68f4d is the unique identifier of the characteristic image of the original image.
  • the feature image corresponding to the original image d3e616c8-1d39-4fd6-a2fe-e6a7e201732d is 6ff55da4-5454-4b81-a831-a25c7bf68f4d.
  • the image description of the characteristic image is "increased and thickened markings of the lower lungs”, the diagnosis result is “bronchitis of the lower lungs”, and the corresponding disease is "bronchitis”.
  • bronchitis The etiology of bronchitis is "viral infection” and/or “cold air”, the symptoms are “cough” and/or “wheezing", and the treatment plan is "taking antiviral, antitussive test drugs", and the drug is "Dalphineum” and / or “Ambroxol Oral Liquid", wherein "taking antiviral and antitussive test drugs” includes the drugs "Dalphineum” and/or "Ambroxol Oral Liquid".
  • the data processing system 110 determines the graph structure of the domain knowledge graph; obtains the original text and original image of the preset domain, extracts multiple entity data from the original text and original image, and determines the relationship between multiple entity data.
  • the association relationship ; build a domain knowledge graph according to the graph structure, multiple entity data and the association relationship between multiple entity data.
  • the knowledge graph usually only contains text; however, the embodiment of the present application establishes a multimodal domain knowledge graph based on text and images. Compared with the prior art, the domain knowledge graph established by the embodiments of the present application is more comprehensive and accurate.
  • the above-mentioned steps of extracting multiple entity data from the original text and original image, and determining the association relationship between the multiple entity data may be implemented in the following manner.
  • One of the entity data extraction methods includes: for the original text, inputting the original text into a pre-trained word segmentation model to obtain multiple word segmentation results output by the word segmentation model; inputting multiple word segmentation results into the pre-trained entity recognition model,
  • the feature text output by the entity recognition model is obtained, and the feature text is determined as entity data.
  • the original text includes image descriptions, diagnosis results, and disease types.
  • the original text can be input into the pre-trained word segmentation model and entity recognition model to obtain feature texts corresponding to image descriptions, diagnosis results, and disease types in the original text.
  • the feature text may include the description of the attributes of the text entity in Table 1.
  • the word segmentation model and the entity recognition model may be trained by the terminal itself, or may be trained by the server, and the terminal may obtain it from the server. This embodiment of the present application does not limit this.
  • the training process of the word segmentation model may include: extracting common words and their standard expressions from massive texts by means of hot word mining; generating synonyms according to the verification results of common words and their standard expressions by verifiers surface.
  • the thesaurus table is converted into a word segmentation dictionary, and the word segmentation dictionary and the manually labeled word segmentation data are used as input to train the word segmentation model.
  • the data processing system 110 receives the verification information input by the verification personnel for the K N-Gram results with higher frequency, determines whether each N-Gram result needs to be combined into a vocabulary according to the verification information, and then combines the results to obtain The vocabulary is updated to the above word segmentation dictionary.
  • an original text "The renal arcuate artery sends out radial branches at regular intervals” is segmented by the word segmentation model and the word segmentation result is "The renal arcuate artery sends out radial branches at regular intervals", from which 9 3-grams can be obtained ⁇ "renal arcuate arteries", “arctic arteries with”, “arteries with regular”, “regular spacing”, “spacing emitting”, “spacing radiating", “radiating”, “radial”, “radiating” branch” ⁇ , artificially confirmed that "renal arcuate artery” is a professional term, so “renal arcuate artery” was merged into “renal arcuate artery” and added to the word segmentation dictionary.
  • the word segmentation model can be retrained according to the updated word segmentation dictionary.
  • the training process of the entity recognition model may include: acquiring a training sample set, and training the recognition model based on the training sample set to obtain the entity recognition model.
  • the above entity recognition model can adopt the Word2Vec+BiLSTM+CRF structure. This embodiment of the present application does not limit the structure of the entity recognition model.
  • the method for determining the association relationship corresponding to the above-mentioned entity data extraction method includes: inputting the original text and multiple feature texts into a pre-trained relationship extraction model, and obtaining the relationship between the multiple feature texts output by the relationship extraction model. association relationship.
  • the relationship extraction model may be trained by the terminal itself; it may also be trained by the server and then acquired by the terminal from the server. This embodiment of the present application does not limit this.
  • the data processing system 110 may input the original text and the plurality of feature texts extracted above into the relationship extraction model, and the relationship extraction model determines the association relationship between the plurality of feature texts according to the content of the original text, and Output the relationship.
  • the above-mentioned relationship extraction model may adopt a gated recurrent unit (Gated Recurrent Unit, GRU) model, and the embodiment of the present application does not limit the structure of the relationship extraction model.
  • GRU Gated recurrent Unit
  • the data processing system 110 can quickly and accurately extract the text entity data from the original text by using the pre-trained word segmentation model, entity recognition model and relationship extraction model, and determine the relationship between the text entity data, It provides a lot of textual basis for constructing domain knowledge graph.
  • Another entity data extraction method includes: for the original image, inputting the original image into a pre-trained recognition model to obtain the recognition result output by the recognition model; cutting out the characteristic image of the region of interest from the original image according to the recognition result, and Identify feature images as entity data.
  • the recognition result is used to indicate the region of interest in the original image.
  • the recognition model adopts a neural network model. After inputting the original image into the neural network model, the neural network model outputs the original image marked with a region of interest recognition frame. Next, the data processing system 110 extracts the feature image in the ROI identification frame as entity data. This embodiment of the present application does not limit the structure of the recognition model.
  • the method for determining the association relationship corresponding to the above entity data extraction method includes: storing the feature image in a feature image library, and storing the association relationship between the feature image and the original image in a preset data table .
  • the feature image is stored in a preset feature image library, and an association relationship such as "original image-feature image" is established. After that, store the relationship in the MySQL data table.
  • This embodiment of the present application does not limit the structure of the data table.
  • the data processing system 110 can quickly and accurately cut out the image entity data from the original image by using the pre-trained recognition model, and determine the relationship between the image entity data and the original image, so as to construct a domain knowledge graph. Provides a lot of image evidence.
  • Another entity data extraction method extract at least one of the texts such as image description, diagnosis result, disease, inducement, symptom, drug, treatment plan, etc. corresponding to the original image from the original text, and determine it as entity data; Feature images are extracted from the images, and the feature images are determined as entity data.
  • the above characteristic images, image description texts and diagnosis results are all entity data.
  • the manner of determining the association relationship corresponding to the foregoing entity data extraction manner may include: establishing an association relationship between the feature image, the image description text, and the diagnosis result text.
  • the data processing system 110 may store the image description text and the diagnosis result text in a MySQL data table, save the characteristic image in the characteristic image database, and store the saving path in the MySQL data table.
  • a unique identifier of the characteristic image is generated, and the unique identifier of the characteristic image is used when establishing the association relationship between the characteristic image and the image description text and the diagnosis result text.
  • the data processing system 110 extracts entity data from the original text and original images, and provides a large amount of textual and image evidences for constructing the domain knowledge graph, making the constructed domain knowledge graph more comprehensive and accurate.
  • the data processing system 110 may establish image-text correspondences based on historical image reports.
  • the data processing system 110 can automatically generate an image report according to the target image and the image-text correspondence.
  • the data processing system 110 may identify the historical image reports (eg, using an image recognition model), obtain characteristic images of the images in the historical image reports, and store the characteristic images in a characteristic image library.
  • the data processing system 110 establishes the corresponding relationship between the feature image and the historical image report, and stores it in the database.
  • the data processing system 110 may search for the second feature image matching the first feature image in the feature image library, and then find the second feature image matching the second feature image through the corresponding relationship between the feature image and the historical image report.
  • the historical image report corresponding to the image Since the template used for the image report is the same, the data processing system 110 can generate the image report of the target image by placing the target image and the basic patient information corresponding to the target image in the corresponding position of the historical image report for replacement.
  • process 500 may be performed by user terminal 130, eg, by CPU 340 and/or GPU 330, and/or by one or more of the modules shown in FIG. 11 .
  • FIG. 8 is an exemplary flowchart of a process of finding image description information and diagnosis result information corresponding to a target image according to some embodiments of the present application.
  • process 800 may be performed by medical system 100 (eg, data processing system 110).
  • process 800 may be performed by a PACS, a post-processing workstation, a physician's workstation, or a server of an imaging system.
  • the process 800 may be performed by the processor 210 .
  • process 800 may be performed by one or more of the modules shown in FIG. 11 .
  • process 800 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 800 may be implemented.
  • a storage device eg, storage device 140
  • the schematic diagram of the operation of process 800 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order of operations of flow 800 illustrated in FIG. 8 and described below is not intended to be limiting.
  • the above-mentioned steps of searching for image description information and diagnosis result information corresponding to the target image based on the pre-built domain knowledge map and the second feature image may include the following steps.
  • the data processing system 110 may search for the second characteristic image in the domain knowledge graph, and obtain an image description set and a diagnosis result set corresponding to the second characteristic image.
  • the image description set may be composed of a plurality of image description texts
  • the diagnosis result set may be composed of a plurality of diagnosis result texts
  • the data processing system 110 extracts a first feature image from the target image, searches a preset feature image library according to the first feature image, and obtains a second feature image matching the first feature image, Wherein, the second characteristic image may be one or more than one. After that, the data processing system 110 searches the domain knowledge graph for the second feature image, and the image description text and diagnosis result text corresponding to each second feature image may be one or more, so that multiple images are acquired Description text and diagnostic result text. Next, a plurality of image description texts are formed into an image description set, and a plurality of diagnosis result texts are formed into a diagnosis result set.
  • the data processing system 110 can perform deduplication processing on the image description set and the diagnosis result set, and obtain the deduplicated image description text and the diagnosis result text, and according to the deduplicated image description
  • the text and diagnosis text determine the image description information and the diagnosis information.
  • the image description set since the image description set includes multiple image description texts, the image description texts may have duplicate descriptions, so the image description set needs to be deduplicated. Similarly, the combination of diagnostic results also needs to be deduplicated.
  • the process of deduplication processing may include: performing word segmentation on an image description text to obtain a word segmentation result; judging whether the word segmentation result contains the text content of other image description texts, and if it contains the content of another image description text, describe the other image description Text removed.
  • the deduplicated image description text and the diagnosis result text are obtained, and the deduplicated image description text and the diagnosis result text are arranged in a certain order to obtain the image description information and the diagnosis result. information.
  • the terminal searches the domain knowledge graph for the second feature image, and obtains the corresponding second feature image.
  • image description set and diagnosis result set ; perform deduplication processing on the image description set and diagnosis result set to obtain the deduplicated image description text and diagnosis result text, and determine the image according to the deduplicated image description text and diagnosis result text
  • Descriptive information and diagnostic result information the terminal uses the domain knowledge image to search for image description information and diagnosis result information. Since the domain knowledge graph is a multimodal knowledge graph including text and images, more comprehensive and accurate images can be found. Descriptive information and diagnostic result information. In addition, the terminal performs deduplication processing on the found image description set and diagnosis result set, which can avoid repeated descriptions in the image report and make the image report more concise.
  • process 800 may be performed by user terminal 130, eg, by CPU 340 and/or GPU 330, and/or by one or more of the modules shown in FIG. 11 .
  • FIG. 9 is an exemplary flowchart of generating an image report according to some embodiments of the present application.
  • process 900 may be performed by medical system 100 (eg, data processing system 110).
  • process 900 may be performed by a PACS, a post-processing workstation, a physician's workstation, or a server of an imaging system.
  • the process 900 may be performed by the processor 210 .
  • the process 900 may be performed by one or more of the modules shown in FIG. 11 .
  • process 900 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 900 may be implemented.
  • a storage device eg, storage device 140
  • process 900 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order of operations of flow 900 illustrated in FIG. 9 and described below is not intended to be limiting.
  • Step 910 the data processing system 110 (for example, the information search module 1130) can search for disease information, inducement information, symptom information, drug information, treatment corresponding to the second characteristic image based on the pre-built domain knowledge graph and the second characteristic image. Program information.
  • the domain knowledge graph associated with the second feature image may also include disease information, and through the disease information, disease-related inducement information, symptom information, treatment plan information, and drug information can be obtained through association query inference, such as shown in Figure 7. Therefore, after finding the image description text and diagnosis result associated with the second feature image in the domain knowledge graph, you can further find the disease information, inducement information, symptom information, treatment plan information and drug information associated with the second feature image. Wait. It can be understood that the disease information, cause information, symptom information, treatment plan information and drug information associated with the second characteristic image can be used as disease information, cause information, symptom information, treatment plan information and drug information corresponding to the target image.
  • the data processing system 110 may generate an image report according to the target image, image description information, diagnosis result information, disease information, cause information, symptom information, treatment plan information and drug information.
  • the data processing system 110 may preset an image report template, and after acquiring the disease information, inducement information, symptom information, treatment plan information and drug information corresponding to the target image, fill these information into the image report template. In the corresponding location, an image report is automatically generated.
  • the data processing system 110 may acquire an image report template.
  • the image report template may include a patient basic information module 1010 , an image description module 1020 , and a diagnosis result module 1030 .
  • the data processing system 110 can fill the information into the corresponding modules of the image report template to automatically generate an image report, such as the image report 1000 shown in FIG. 10 .
  • the data processing system 110 searches for disease information, inducement information, symptom information, treatment plan information and drug information corresponding to the target image based on the pre-built domain knowledge map and the second feature image; Information, diagnosis result information, disease information, predisposing information, symptom information, treatment plan information, and drug information generate image reports.
  • the domain knowledge graph includes not only feature images, image description texts, and diagnosis result texts, but also disease information, inducement information, symptom information, treatment plan information, and drug information. Therefore, compared with the prior art, The embodiments of the present application enrich the content of the image report in combination with the intelligent query and reasoning functions of the knowledge graph, and can provide doctors with more comprehensive information support.
  • process 900 may be performed by user terminal 130, eg, by CPU 340 and/or GPU 330, and/or by one or more of the modules shown in FIG. 11 .
  • FIG. 11 is an exemplary block diagram of an image report generating apparatus 1100 according to some embodiments of the present application.
  • the hardware and/or software of the data processing system 110 and/or the user terminal 130 may be implemented based on the image report generating apparatus 1100 .
  • the image report generation apparatus 1100 may include an image capture module 1110 , an image search module 1120 , an information search module 1130 , and a report generation module 1140 .
  • the image capturing module 1110 is configured to acquire a target image, and capture a first characteristic image from the target image, wherein the first characteristic image is an image of a region of interest in the target image.
  • the image search module 1120 is configured to search in a preset feature image library according to the first feature image, and obtain a second feature image matching the first feature image.
  • the information search module 1130 is used to search for text information corresponding to the second characteristic image based on the pre-built domain knowledge map and the second characteristic image, for example, image description information, diagnosis result information, disease, incentive, symptom, drug, At least one of the treatment regimens and the like.
  • the information search module 1130 may include a first information search module, configured to search for image description information and diagnosis result information corresponding to the second characteristic image based on the pre-built domain knowledge graph and the second characteristic image.
  • the report generating module 1140 is configured to generate an image report of the target image.
  • the report generating module 1140 may include a first report generating module 1140 for generating an image report according to the target image, image description information and diagnosis result information.
  • the information search module 1130 further includes:
  • the structure determination module is used to determine the graph structure of the domain knowledge graph; the graph structure includes entity types and entity connection relationships;
  • the data and relationship extraction module is used to obtain the original text and original image in the preset field, extract multiple entity data from the original text and original image, and determine the association relationship between multiple entity data;
  • the graph building module is used to build a domain knowledge graph according to the graph structure, multiple entity data, and the relationship between multiple entity data.
  • the entity type includes at least one of raw image, feature image, image description, diagnosis, disease, cause, symptom, drug, treatment regimen.
  • the entity connection relationship includes at least one of observe, inhibit, juxtapose, select, use, characterize, result, relate, complication, contain, induce, manifest, and the like.
  • the above-mentioned data and relationship extraction module is specifically used to input the original text into a pre-trained word segmentation model to obtain multiple word segmentation results output by the word segmentation model;
  • the trained entity recognition model the feature text output by the entity recognition model is obtained, and the feature text is determined as entity data.
  • the above-mentioned data and relationship extraction module is specifically configured to input the original text and multiple feature texts into a pre-trained relationship extraction model to obtain the association relationship between multiple feature texts output by the relationship extraction model.
  • the above-mentioned data and relationship extraction module is specifically used to input the original image into a pre-trained recognition model to obtain a recognition result output by the recognition model; the recognition result is used to indicate the sense of sensation in the original image. Area of interest; the characteristic image of the area of interest is cut out from the original image, and the characteristic image is determined as entity data.
  • the image report generating apparatus 1100 further includes:
  • the image storage module 2730 is configured to store the characteristic image in the characteristic image library, and store the relationship between the characteristic image and the original image in a preset data table.
  • the above data and relationship extraction module is specifically configured to extract image description text and diagnosis result text corresponding to the original image from the original text, and determine the image description text and diagnosis result text as entity data; Feature images are extracted from the images, and the feature images are determined as entity data; the relationship between the feature images, image description texts and diagnosis result texts is established.
  • the above-mentioned graph building module is specifically configured to form a plurality of entity data into a plurality of triples according to the association relationship between the plurality of entity data; and store the plurality of triples in a preset graph database , get the domain knowledge graph.
  • the above-mentioned first information search module is specifically configured to search for the second feature image in the domain knowledge graph, and obtain an image description set and a diagnosis result set corresponding to the second feature image;
  • the image description set consists of multiple images It consists of description text, and the diagnosis result set consists of multiple diagnosis result texts;
  • the image description set and the diagnosis result set are deduplicated to obtain the deduplicated image description text and the diagnosis result text, and according to the deduplicated image description text and diagnosis result text to determine image description information and diagnosis result information.
  • the above-mentioned image search module 1120 is specifically configured to perform similarity calculation between the first feature image and each feature image in the feature image library to obtain multiple similarities;
  • the characteristic image of the set condition is determined as the second characteristic image.
  • the information search module 1130 further includes a second information search module, configured to search for disease information, inducement information, symptom information, and treatment plan corresponding to the target image based on the pre-built domain knowledge graph and the second feature image. information and medication information.
  • the report generation module 1140 further includes a second report generation module for generating an image report according to the target image, image description information, diagnosis result information, disease information, predisposing information, symptom information, treatment plan information and drug information .
  • each module in the above-mentioned image report generating apparatus 1100 may be implemented in whole or in part by software, hardware, or a combination thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • the present application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor (eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU) 330, or one or more modules of FIG. 11) when executed, implements at least one of the process 400, the process 500, the process 800, and the process 900.
  • a processor eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU
  • the present application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor (eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU) 330, or one or more modules of FIG. 11) when executed, implements at least one of the process 400, the process 500, the process 800, and the process 900.
  • a processor eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU
  • FIG. 12A is an exemplary flowchart of a text error correction method according to some embodiments of the present application.
  • process 1200 may be performed by medical system 100 (eg, data processing system 110).
  • the process 1200 may consist of a Image Archiving and Transmission System (PACS), Post-processing Workstation, Surgical System, Hospital Information System (HIS), Laboratory Information System (LIS), Radiology Information System (RIS), Doctor Workstation, Nurse Workstation , imaging system, hospitalization management system, outpatient and emergency management system, or server implementation of charging system.
  • the process 1200 may be performed by the processor 210 .
  • process 1200 may be performed by one or more of the modules shown in FIG. 22 .
  • process 1200 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 1200 may be implemented.
  • a storage device eg, storage device 140
  • the schematic diagram of the operation of process 1200 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order of operations of flow 1200 illustrated in FIG. 12A and described below is not intended to be limiting.
  • the data processing system 110 may perform error detection on the text to be processed to obtain the type and location of the erroneous text.
  • the text to be processed is text awaiting error detection and correction.
  • Computer equipment can obtain pending texts in medical-related systems such as Hospital Information System (HIS), Radiology Information System (RIS), electronic medical record system, speech recognition result text system, question answering system, etc. , of course, the computer equipment can also obtain the text to be processed from the systems in other fields.
  • Types of erroneous text include typo types, missing types, out-of-order types, and redundant types.
  • the position of the erroneous text indicates the position of the erroneous text in the text to be processed.
  • the text may be at least one of symbols, numbers, letters, single words, words, and sentences.
  • Error detection may include detecting whether erroneous text is included in the text to be processed, and/or determining the type and location of erroneous text.
  • the computer device for example, the data processing system 110
  • the text to be processed may be input into the trained error detection model for error detection, and the data contained in the text to be processed can be obtained.
  • the type of error text and the position of the error text in the text to be processed may use a neural network model or a machine learning model.
  • the BiLSTM-CRF model is widely used because of its good error detection function, and the selection of the detection model is not limited in this embodiment.
  • this embodiment also provides a method for training the above-mentioned BiLSTM-CRF model, the method comprising: using a preset sequence labeling method to label the error type and position of the error text in the sample text, and obtain the labelled text. , and then input the labeled text as a training sample into the initial BiLSTM-CRF model for training, and obtain the trained BiLSTM-CRF model, so that the trained BiLSTM-CRF model can output the position of the wrong text according to the input pending text. and type for the above usage.
  • the above sequence labeling method may adopt any type of labeling method, for example, the Bio sequence labeling method or the Bioes sequence labeling method.
  • sample text is "This material shows how much damage smoking has caused people, but the wife in the article is still conscious, and she still has a meaning to live"
  • enter the text to be processed Go to the Bio sequence labeling model for labeling
  • the labelled sample text is "this/O species/O material/O material/O table/BS indicated/IS smoke/O smoke/O cited/O started/O came/O people /BRs/Too many IRs/O what/O big/O's/O damage/O harm/O but/O is/O text/O chapter/O middle/O's/O wife/O son/O is/BW
  • /IW has /O meaning /O awareness /O of /O
  • she /O also /O has /O live /BW /IW meaning /O meaning /O”
  • R represents redundant type text
  • S represents error type text
  • W represents unordered type text
  • B represents the beginning of
  • the data processing system 110 may determine a target error correction strategy corresponding to the type of the error text according to the correspondence between the error text type and the error correction strategy.
  • the error correction strategy represents the method by which erroneous text is corrected for errors.
  • the correspondence between error text types and error correction strategies can be pre-stored in the database for later review and use.
  • the computer device when it obtains the type of the erroneous text based on the foregoing steps, it can further determine the type of the erroneous text according to the type of the erroneous text and the correspondence between the type of the erroneous text and the error correction strategy.
  • a target error correction strategy so that the corresponding type of erroneous texts are subjected to error correction processing according to the error correction method indicated by the target error correction strategy.
  • the corresponding target error correction strategy includes a phonetic comparison error correction strategy and/or a similarity comparison error correction strategy; if the type of the error text is a missing type or a disordered type, the corresponding The target error correction strategy includes a matching degree comparison error correction strategy; if the type of the error text is a redundant type, the corresponding target error correction strategy includes a deletion error correction strategy.
  • the data processing system 110 may perform error correction processing on the erroneous text according to the target error correction strategy and the position of the erroneous text.
  • the computer device after the computer device determines the target error correction strategy corresponding to the type of the erroneous text, it can find the error correction method in the text to be processed according to the position of the error text according to the error correction method indicated by the target error correction strategy. Error text, or find the text related to the error text, and then perform different error correction operations such as adding, deleting, or correcting the error text or the text related to the error text, so as to obtain the corrected text.
  • the type and location of the erroneous text are obtained by performing error detection on the text to be processed, and the type corresponding to the erroneous text is determined according to the correspondence between the type of the erroneous text and the error correction strategy.
  • the target error correction strategy is further used to correct the error text according to the target error correction strategy and the position of the error text.
  • the error correction operation for each type of error text is realized, and compared with the traditional error text regardless of any type, the error correction operation is realized.
  • the use of a large database to obtain texts for error correction leads to the problems of large data processing volume and poor error correction accuracy.
  • the error correction method described in this application greatly improves the error correction accuracy and data processing speed.
  • an implementation manner of the above step 1230 is provided, as described in FIG. 13 , the above step 1230 "according to The target error correction strategy and the location of the erroneous text, and the error correction processing of the erroneous text", including:
  • the data processing system 110 may determine at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text.
  • the computer device may determine at least one candidate text corresponding to the erroneous text by comparing the erroneous text with the pinyin of a plurality of preset texts. In some embodiments, the computer The device may also determine at least one candidate text corresponding to the erroneous text by comparing the similarity between the erroneous text and a plurality of preset texts. When the type of the erroneous text includes a missing type or an out-of-order type, the computer device may determine at least one candidate text corresponding to the erroneous text by comparing the matching degrees of the erroneous text with a plurality of preset texts.
  • the similarity or matching degree between the error text and the preset text may be determined according to at least one of pronunciation, spelling, text, semantics, context, etc. of the error text and the preset text.
  • preset texts whose similarity or matching degree is greater than a threshold eg, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, etc. may be determined as candidates text.
  • the data processing system 110 may perform error correction processing on the erroneous text according to the at least one candidate text.
  • the computer device After the computer device obtains at least one candidate text based on the above steps, it can arbitrarily select one candidate text from the multiple candidate texts to correct the erroneous text. Select a candidate text from the candidate texts to correct the error text. When correcting the error, the selected candidate text can be used to replace the error text, or the selected candidate text can be added to the corresponding position of the text to be processed.
  • the above-mentioned preset rules may be determined in advance by computer equipment according to application requirements. For example, the preset rules may include selecting candidate texts with the highest similarity or matching degree.
  • the above error correction data processing method realizes error correction for typo type, missing type, and disorder type error text, and each type of error text corresponds to a different target error correction strategy, so that the computer equipment in the error correction process can be corrected.
  • Corresponding error correction operations can be performed according to specific types of error texts, which greatly improves error correction accuracy.
  • the above-mentioned step 1310 “determines at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text”, include:
  • Step 1410 Extract the typo-type text from the to-be-processed text according to the position of the typo-type text.
  • the typo-type text can be found in the text to be processed according to the position of the typo-type text, and the typo-type text can be found in the text to be processed. Extract it for later retrieval using that typo type of text.
  • Step 1420 Select from the preset first database the text that is homophonic to the text of the typo type, and/or that has a similarity with the text of the typo type that is greater than a preset similarity threshold, as candidate text.
  • the first database may comprise any type of database.
  • the preset similarity threshold may be determined in advance by the computer device according to the error correction accuracy. Specifically, when the computer device obtains the typo-type text, the typo-type text can be further used as the target text to search the first database for the text with the same pronunciation as the target text. In some embodiments, the computer device can also be Search the first database for texts that have the same homophony as the target text, and then use the retrieved texts as candidate texts; in some embodiments, the computer device may also calculate the similarity between the target text and each text in the first database , and then use the text whose similarity is greater than the preset similarity threshold as candidate text.
  • the computer device may also calculate the edit distance between the target text and each text in the first database, and then use the text whose edit distance is less than a preset edit distance threshold as candidate text, where the preset edit distance threshold It can be determined in advance by computer equipment according to the error correction accuracy.
  • the above-mentioned method for determining a candidate text can determine the candidate text corresponding to the text of the typo type by comparing the pinyin of the wrong text, or the similarity between the wrong text and each text in the first database.
  • the error correction speed can be improved when the candidate text is processed for error correction.
  • the above-mentioned step 1310 “determines at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text”, include:
  • Step 1510 according to the position of the missing type text, extract the text adjacent to the missing type text from the text to be processed.
  • the text adjacent to the text of the missing type can be found in the text to be processed according to the position of the text of the missing type, and the adjacent text can be extracted out for later retrieval using that adjacent text.
  • Step 1520 Select texts matching the adjacent texts from the preset second database as candidate texts.
  • the second database may include a database dedicated to the medical field, for example, a medical knowledge graph database.
  • the second database in this embodiment may be the same or different from the first database in the foregoing embodiments.
  • the adjacent text can be further used as the target text in the second database according to at least one of pronunciation, spelling, text, semantics, context, etc. retrieve text that matches this target text, and use the retrieved text as candidate text. For example, if the to-be-processed text is "Hospital Discharge Department Workflow", then the detected missing type text is "Dispatch Department”, and the text adjacent to the missing type text is "Hospital XX Workflow”. Take “hospital XX workflow” as the target text to search in the second database, and obtain “hospital radiology workflow”, then the text matching the missing type of text is "radiology”.
  • the above-mentioned method for determining candidate texts can determine the candidate texts corresponding to the missing types of erroneous texts by comparing the degree of matching between the erroneous texts and each text in the first database.
  • the method is simple and practical, and enables subsequent error correction based on the candidate texts.
  • the error correction speed can be increased during processing.
  • the above-mentioned step 1310 “determines at least one candidate text corresponding to the erroneous text according to the target error correction strategy and the position of the erroneous text”. ,include:
  • Step 1610 Extract the text of the disordered type from the text to be processed according to the position of the text of the disordered type.
  • the unordered type of text when the type of error text detected by the computer device is an unordered type, the unordered type of text can be found in the text to be processed according to the position of the unordered type of text, and the unordered type of text can be found in the text to be processed.
  • the text of the order type is extracted, so that the text of the unorder type can be used for retrieval later.
  • Step 1620 Select texts containing the same characters as the disordered texts from the preset third database as candidate texts.
  • the third database may include a database dedicated to the medical field, for example, a medical knowledge graph database.
  • the third database in this embodiment may be the same as or different from the second database in the foregoing embodiment.
  • the computer device when it obtains the text of the disordered type, it can further use the text of the disordered type as the target text to retrieve the text containing the same words as the target text in the third database, and then use the retrieved text as the target text.
  • candidate text For example, if the text to be processed is "Workflow of Radiology in a Hospital", the detected text of the disordered type is "Radiology”, and the text containing the same words as the text of the disordered type is "Radiology”.
  • the above method for determining candidate texts can determine the candidate texts corresponding to the disordered type of error texts by searching the third database for each text that contains the same word as the error text.
  • the error correction speed can be increased during error correction processing.
  • the present application also provides an implementation manner of the foregoing 1230, as shown in FIG. 17 , the manner includes:
  • Step 1710 Add at least one candidate text to a corresponding position in the text to be processed in sequence, and score the smoothness of the added text to be processed.
  • the multiple candidate texts can be further added to corresponding positions in the text to be processed in sequence, and after multiple additions are obtained pending text. Then the computer device can sequentially input multiple added texts to be processed into the trained language model to score the degree of smoothness.
  • a high score means that the added text to be processed is very smooth, and a low score means the added text to be processed. Text is not smooth.
  • the above language model is used to analyze the smoothness of the added text to be processed, and to score the added text to be processed according to the level of smoothness.
  • training sample data can be obtained from a medical professional database, and then the training sample data can be input into a preset pre-trained language model for training to obtain a trained language model.
  • the pre-trained language model may use the existing BERT pre-trained language model, or other types of pre-trained language models, which are not limited here.
  • Step 1720 Sort the multiple scored texts to be processed in descending order of scores, and determine the text to be processed with the highest score as the corrected text.
  • the computer device After the computer device scores the multiple added texts to be processed, it can further sort the multiple scored texts to be processed in descending order of the scores, and then select the text to be processed with the highest score as the Corrected text; in some embodiments, the computer device may first select the text to be processed whose score is greater than the preset score threshold from a plurality of scored texts to be processed, if the text to be processed whose score is greater than the preset score threshold When the number is greater than one, you can further select any text to be processed from the selected multiple texts to be processed, or select a text to be processed according to the preset rules, and finally select the text to be processed by the above method.
  • the text to be processed is the corrected text.
  • the trained language model is used to score the smoothness of the added text to be processed.
  • the method described in this example improves the accuracy of selecting the corresponding candidate text for error correction according to the score because the context information of the erroneous text is combined for scoring.
  • the above step 1230 “according to the target error correction strategy and the position of the error text, correct the Error text for error correction”, including:
  • Step 1810 determine whether to delete the redundant type of text, if it is determined to delete, go to step 1820, if it is determined not to delete, go to step 1830.
  • the computer device may input the redundant type of text into the trained language model for detection, and detect whether the redundant type of text is fluent, in some embodiments , the computer device can also extract the text containing the redundant type of text from the text to be processed, and then input the extracted text into the trained language model for detection to detect whether the extracted text is smooth, if smooth It means that the text of this redundant type is actually not redundant text in the text to be processed, and if it is inconsistent, it means that the text of this redundant type is actually redundant text in the text to be processed.
  • the trained language model may output a result of "smooth” or “unsmooth”, for example, the trained language model may output "1" for “smooth” and output "0" for “unsmooth” .
  • the trained language model may output a numerical value representing the fluent degree, for example, the trained language model may output "90%” indicating that the fluent degree is 90%. Texts with a fluent degree greater than or equal to a certain threshold (for example, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, etc.) can be considered fluent, and texts with a fluent degree less than the threshold can be considered as fluent Not fluent.
  • a certain threshold for example, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, etc.
  • the above language model is used to analyze whether the text of redundant type or the text containing the text of redundant type is smooth, and output the analysis result including smooth or non-smooth.
  • training sample data can be obtained from a medical professional database, and then the training sample data can be input into a preset pre-trained language model for training to obtain a trained language model.
  • the pre-trained language model may use the existing BERT pre-trained language model, or other types of pre-trained language models, which are not limited here.
  • Step 1820 delete the wrong text in the text to be processed.
  • This embodiment relates to an application scenario in which a computer device determines to delete redundant texts.
  • the computer device directly performs an operation of deleting redundant text in the text to be processed.
  • Step 1830 do not perform the delete operation.
  • This embodiment relates to an application scenario in which the computer device determines not to delete redundant texts.
  • the computer device does not perform the delete operation.
  • the above-mentioned 1810 “determine whether to delete redundant text according to the target error correction strategy and the position of the erroneous text” includes:
  • Step 1910 Extract the text containing the redundant type of text from the to-be-processed text according to the position of the redundant type of text.
  • the redundant type of text when the type of error text detected by the computer device is a redundant type, the redundant type of text can be found in the text to be processed according to the position of the redundant type of text, and the redundant type of text can be found in the text to be processed.
  • the text of the redundant type of text is extracted for later retrieval using the text containing the redundant type of text. For example, if the text to be processed is "Medical Radiology Workflow", and the redundant type of text is "Release”, then the extracted text containing the redundant type of text can be "Release and Radiology", or It can be "medical radiology department", and the length of the text including redundant text can be predetermined by the computer equipment, which is not limited here.
  • Step 1920 Detect whether the text including the redundant type of text is smooth, if not, go to Step 1930, and if it is smooth, go to Step 1940.
  • Step 1930 determining to delete redundant text.
  • Step 1940 determining not to delete redundant texts.
  • the text when the computer device obtains text containing redundant types of text, the text can be input into the trained language model for detection to detect whether the extracted text is smooth, if not, then It is determined to delete the redundant type of text, and if it is fluent, it is determined not to delete the redundant type of text.
  • the trained language model is used to analyze the text containing the redundant type of text to judge whether the text is smooth, compared to the judgment caused by only analyzing the redundant type of text.
  • the method described in this embodiment combines the context information of the redundant type of text to determine whether it needs to be deleted, which improves the accuracy of error correction for the redundant type of text.
  • the present application also provides a data processing method, as shown in FIG. 20 , the data processing method includes:
  • Step 2001 Perform error detection on the text to be processed to obtain the type and location of the erroneous text.
  • Step 2003 judging the type of error text.
  • the type of error text includes typo types, execute steps 2004-2005 and 2010-2011; when the type of error text includes missing types, execute steps 2006-2007 and 2010-2011; when the type of error text includes disordered types , perform steps 2008-2009 and 2010-2011; when the type of error text includes redundant types, perform steps 2012-2015.
  • Step 2004 Extract typo-type text from the to-be-processed text according to the position of the typo-type text.
  • Step 2005 from the preset first database, select texts that are homophones with typo-type texts, and/or have a similarity with typo-type texts greater than a preset similarity threshold as candidate texts.
  • Step 2006 according to the position of the missing type text, extract the text adjacent to the missing type text from the text to be processed.
  • Step 2007 Select texts matching the adjacent texts from the preset second database as candidate texts.
  • Step 2008 Extract the text of the disordered type from the text to be processed according to the position of the text of the disordered type.
  • Step 2009 Select from the preset third database the text containing the same characters as the disordered text as the candidate text.
  • Step 2010 Add at least one candidate text to a corresponding position in the text to be processed in sequence, and score the smoothness of the added text to be processed.
  • Step 2011 sort the plurality of scored texts to be processed in descending order of scores, and determine the text to be processed with the highest score as the corrected text.
  • Step 2012 according to the position of the redundant type of text, extract the text containing the redundant type of text from the to-be-processed text.
  • Step 2013 Detect whether the text including the redundant type of text is fluent, if not, go to Step 2014, and if it is fluent, go to Step 2015.
  • step 2014 it is determined to delete redundant texts, and the erroneous texts in the to-be-processed texts are deleted.
  • Step 2015 it is determined that the redundant type of text is not to be deleted, and the deletion operation is not performed.
  • the above description is only for illustration and description, and does not limit the scope of application of the present specification. Various modifications and changes can be made by those skilled in the art under the guidance of this specification. However, these corrections and changes are still within the scope of this specification.
  • the above process may be performed by the user terminal 130, for example, the CPU 340 and/or the GPU 330, and/or one or more modules shown in FIG. 22.
  • the present application also provides an error correction system 2100 , as shown in FIG. 21 , the error correction system includes: an error detection model 2110 , a screening and recall model 2120 , and an error correction processing model 2130 .
  • the output end of the error detection model 2110 is connected with the input end of the screening and recall model 2120
  • the output end of the screening and recall model 2120 is connected with the input end of the error correction processing model 2130 .
  • the error detection model 2110 is used to detect the error text on the input text, and output the type and position of the error text; when the type of the error text includes at least one of typo type, missing type, and disorder type
  • the screening recall model 2120 is used to select an error correction processing strategy corresponding to the type of error text output by the error detection model, and then according to the selected error correction processing strategy and the position of the error text, in the preset database Determine multiple candidate texts corresponding to the erroneous text, and then the error correction processing model scores the smoothness of the multiple candidate texts, and corrects the text to be processed according to the candidate text with the highest score; when the type of erroneous text includes redundancy type, the screening recall model 2120 is used to detect redundant types of texts, or whether the texts including redundant types of texts are smooth, to determine whether the redundant types of texts need to be deleted, and then the error correction processing model 2130 is used to screen the recall models.
  • the aforementioned data processing method is applied to the aforementioned error correction system for performing error correction processing on text.
  • the functions and implementation methods of each model or each component included in the error correction system please refer to the description of the aforementioned embodiment, and the description will not be repeated here.
  • the hardware and/or software of data processing system 110 and/or user terminal 130 may be implemented based on error correction system 2100 .
  • the type of error text may also include content errors.
  • an image report can include target images and text. If the content of a piece of text does not correspond to other texts or target images, the text is error text, and the error type is content error.
  • the inspection item in the image report is "CT axial plain scan of the head", and the image description and diagnostic opinion in the image report are descriptions about the thoracic cavity, then, the image description and diagnostic opinion in the image report
  • the text is error text and the error type is content error.
  • the target image in the image report is a head image
  • the image description and diagnostic opinion in the image report are descriptions about the thoracic cavity
  • the image description and diagnostic opinion text in the image report are error text
  • the error type is content error.
  • the image report can be input into a pre-trained word segmentation model, and multiple word segmentation results of the text in the image report output by the word segmentation model can be obtained. Input multiple word segmentation results into a pre-trained text recognition model to determine irrelevant word segmentation in the word segmentation results.
  • the irrelevant participle location sentences recognized by the text recognition model can be determined as erroneous texts, and the type and location of the erroneous texts can be determined.
  • the target image in the image report can be extracted, and the extracted target image can be input into the image recognition model to obtain image features output by the model.
  • the image features and the above multiple word segmentation results are input into the pre-trained text recognition model, and the word segmentation results that are not related to the image features in the word segmentation results are determined.
  • the irrelevant participle location sentences recognized by the text recognition model can be determined as error text, and the type and location of the error text can be determined.
  • the method described in FIG. 4-FIG. 11 can be used to search the knowledge spectrum again for the text description for replacing the wrong text, and update the image report.
  • a data processing apparatus 2200 including:
  • the detection module 2210 is used to perform error detection on the text to be processed to obtain the type and location of the error text;
  • a determination module 2220 is used to determine the target error correction strategy corresponding to the type of the error text according to the correspondence between the error text type and the error correction strategy;
  • the error correction module 2230 is configured to perform error correction processing on the error text according to the target error correction strategy and the position of the error text.
  • Each module in the above-mentioned data processing apparatus may be implemented in whole or in part by software, hardware, and combinations thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • the hardware and/or software of the data processing system 110 and/or the user terminal 130 may be implemented based on the data processing apparatus 2200 .
  • the present application provides a computer-readable storage medium having stored thereon a computer program that is executed by a processor (eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU 330, the error correction system 2100 of FIG. 21, or one or more modules of FIG. 22) implements at least one of the processes 1200-2000 when executed.
  • a processor eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU 330, the error correction system 2100 of FIG. 21, or one or more modules of FIG. 22
  • the storage of unstructured medical data is particularly important.
  • the storage of unstructured medical data is mainly based on the data identification of unstructured medical data, using customized interfaces and backup software plus tape library or virtual tape library for storage.
  • medical data is usually stored in Massive data, when using this method to store unstructured medical data, it takes a long time to scan the file system only once. Therefore, the traditional storage method of unstructured medical data has the problem of low storage efficiency.
  • the present application provides an unstructured data storage method, device, computer equipment and storage medium that can improve the storage efficiency of unstructured medical data.
  • FIG. 23 is an exemplary flowchart of an unstructured data storage method according to some embodiments of the present application.
  • process 2300 may be performed by medical system 100 (eg, data processing system 110).
  • the process 2300 may consist of a Image Archiving and Transmission System (PACS), Post-processing Workstation, Surgical System, Hospital Information System (HIS), Laboratory Information System (LIS), Radiology Information System (RIS), Doctor Workstation, Nurse Workstation , imaging system, hospitalization management system, outpatient and emergency management system, or server implementation of charging system.
  • the process 2300 may be performed by the processor 210 .
  • process 2300 may be performed by one or more of the modules shown in FIG. 27 .
  • process 2300 may be stored in a storage device (eg, storage device 140 ) in the form of programs or instructions, and when medical system 100 executes the program or instructions, process 2300 may be implemented.
  • a storage device eg, storage device 140
  • the operational schematic diagram of process 2300 presented below is illustrative. In some embodiments, this process may be accomplished using one or more additional operations not described and/or one or more operations not discussed. Additionally, the order of operations of flow 2300 illustrated in FIG. 23 and described below is not intended to be limiting.
  • Step 2310 Obtain unstructured medical data to be stored.
  • unstructured data refers to data that cannot be represented in database tables with two-dimensional logic.
  • unstructured medical data may refer to office documents, texts, pictures, XML, HTML, various reports, audio, video, etc., such as Computed Tomography (CT) images, Magnetic Resonance Imaging (MRI) images, PET images, Molecular Imaging (MI), X Unstructured medical image data such as X-ray (X-ray, XR) images, ultrasound images, image reports, diagnostic reports, laboratory tests, application forms, texts, pictures, audios, videos, etc. obtained by the medical system 100 from external devices.
  • CT Computed Tomography
  • MRI Magnetic Resonance Imaging
  • MI Molecular Imaging
  • X Unstructured medical image data such as X-ray (X-ray, XR) images, ultrasound images, image reports, diagnostic reports, laboratory tests, application forms, texts, pictures, audios, videos, etc. obtained by the medical system 100 from external devices.
  • a computer device acquires unstructured medical data to be stored.
  • the computer device may obtain the unstructured medical data to be stored from the components of the medical system 100, for example, obtain the unstructured medical data to be stored from a PACS (Picture Archiving and Communication Systems) server.
  • PACS Picture Archiving and Communication Systems
  • unstructured medical data to be stored can also be obtained in real time from medical imaging equipment.
  • the computer device may obtain the unstructured medical data to be stored from a device external to the medical system 100 .
  • the computer device may first preprocess the acquired unstructured medical data to be stored, and remove any unstructured medical data in the acquired unstructured medical data to be stored. redundant data, etc.
  • Step 2320 parse the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data, and the corresponding relationship (index information) between the unstructured medical data and the structured medical data.
  • Structured data may refer to data that is logically expressed and implemented by a two-dimensional table structure.
  • Unstructured data can be parsed through offline parsing or online parsing. Specifically, the computer equipment parses the acquired unstructured medical data to be stored, and obtains structured medical data corresponding to the unstructured medical data, and the relationship between the unstructured medical data and the obtained structured medical data corresponding relationship.
  • the computer device may use a preset image analysis method to analyze the unstructured medical data to be stored to obtain structured medical data corresponding to the unstructured medical data to be stored, or use a preset text
  • the extraction method analyzes the unstructured medical data to be stored, obtains structured medical data corresponding to the unstructured medical data to be stored, and then obtains unstructured medical data according to the unstructured medical data to be stored and the obtained structured medical data. Correspondence between structured medical data and structured medical data.
  • the computer device may obtain the correspondence between the unstructured medical data to be stored and the structured medical data according to the parsing path when parsing the unstructured medical data to be stored.
  • the parsing methods corresponding to different types of unstructured data are also different, and accordingly, the correspondence between unstructured medical data and structured medical data corresponding to different types of unstructured data is also different.
  • the computer device may determine a parsing method corresponding to the unstructured medical data to be stored according to the type of the unstructured medical data to be stored, so as to parse the unstructured medical data to obtain the Structured medical data corresponding to unstructured medical data.
  • the computer device may further perform maximum set standard processing on the correspondence between the unstructured medical data and the structured medical data corresponding to different types of unstructured data, as needed, that is, the computer device According to the needs of users, the corresponding relationship between unstructured medical data and structured medical data can be processed by the largest set standard, and the corresponding relationship between unstructured medical data and structured medical data can be stored in different sets. .
  • Step 2330 Store the structured medical data and the corresponding relationship in a preset fourth database, and store the unstructured medical data in a preset fifth database.
  • the computer device stores the above-obtained correspondence between the structured medical data and the unstructured medical data and the structured medical data in a preset fourth database, and stores the unstructured data to be stored in the preset fourth database. in the fifth database set up.
  • the preset fourth database may be a DB database
  • the preset fifth database may be an object storage database.
  • the computer device may store the structured medical data and the obtained correspondence between the unstructured medical data and the structured medical data in a preset fourth database in real time, or the corresponding relationship between the obtained unstructured medical data and the structured medical data may be stored in a preset fourth database.
  • the time interval is, for example, 5 seconds, etc.
  • the corresponding relationship between the structured medical data and the obtained unstructured medical data and the structured medical data is stored in the preset fourth database.
  • the computer equipment can The stored unstructured medical data is stored in the preset fifth database, and the unstructured medical data to be stored can also be stored in the preset fifth database according to the preset time interval.
  • the computer equipment can quickly and accurately obtain the structured medical data corresponding to the unstructured medical data to be stored, and the unstructured medical data. In this way, the computer equipment can quickly store the obtained structured medical data and the obtained corresponding relationship in the preset fourth database, and store the unstructured medical data to be stored.
  • the chemical medical data is stored in the preset fifth database, thereby improving the storage efficiency of the unstructured medical data to be stored.
  • the above step 2320 includes:
  • Step 2410 using preset parsing rules to parse the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data.
  • the computer device uses preset parsing rules to parse the unstructured medical data to be stored to obtain structured medical data corresponding to the unstructured medical data to be stored.
  • the preset parsing rule may be a preset text parsing rule, a preset image parsing rule, or a preset report parsing rule.
  • the computer device may use a preset text parsing rule to parse the text in the unstructured data to be stored to obtain structured medical data corresponding to the unstructured data to be stored.
  • Step 2420 Establish a correspondence between the unstructured medical data and the structured medical data according to the identifier of the unstructured medical data and the identifier of the structured medical data.
  • the computer device establishes the correspondence between the unstructured medical data to be stored and the obtained structured medical data according to the identification of the unstructured medical data to be stored and the identification of the obtained structured medical data.
  • the identifier of the unstructured medical data to be stored as A and the identifier of the structured medical data corresponding to the obtained unstructured medical data as B as an example, the created unstructured medical data to be stored and The corresponding relationship between the obtained structured medical data is: AB.
  • the computer equipment adopts the preset parsing rules, and the process of parsing the unstructured medical data to be stored is very simple, so that the computer equipment can quickly obtain the structured medical data corresponding to the unstructured medical data to be stored.
  • Medical data which improves the efficiency of computer equipment in obtaining structured medical data corresponding to unstructured medical data to be stored, thereby enabling computer equipment to obtain structured medical data according to the identification of the unstructured medical data to be stored and the identification of the obtained structured medical data. , quickly establish the correspondence between the unstructured medical data to be stored and the obtained structured medical data, thereby improving the correspondence between the unstructured medical data to be stored and the obtained structured medical data obtained by computer equipment relationship efficiency.
  • the preset parsing rules include: medical image text recognition rules, DICOM file parsing rules, and XML reports Parsing rules.
  • the above-mentioned preset parsing rules include medical image text recognition rules, DICOM file parsing rules, and XML report parsing rules, that is, the computer equipment can use medical image text recognition rules, DICOM file parsing rules, and XML report parsing rules.
  • the stored unstructured medical data is parsed to obtain structured medical data corresponding to the unstructured medical data to be stored.
  • the computer device can use medical image text recognition rules, DICOM file parsing rules and XML report parsing rules to parse the unstructured medical data to be stored in real time, or can use medical image text recognition rules, DICOM file parsing rules After the rules and XML report parsing rules are parsed in the background of the unstructured medical data to be stored, the corresponding structured medical data obtained are transmitted to the computer equipment.
  • the medical image text recognition rule identifies each text in the unstructured medical data to be stored by preconfiguring the meaning represented by each text, parses out the meaning represented by each text, and obtains the unstructured medical data to be stored.
  • the structured medical data corresponding to the data for example, the medical image text recognition rules can parse the application form to obtain the meaning represented by each word in the application form, so as to obtain the structured medical data corresponding to the application form;
  • the DICOM file parsing rules are Medical digital imaging and communication rules, an international standard for medical images and related information (ISO 12052), which defines a medical image format that can be used for data exchange with a quality that meets clinical needs.
  • the medical data is parsed to obtain the corresponding structured medical data; the XML report parsing rules are configured to parse the unstructured medical data to be stored by configuring custom parsing rules, for example, to parse the CDA clinical document structure, in some embodiments , CDA provides a general schema that can express all possible documents, all CDA documents are expressed in XML encoding, therefore, CDA documents can be parsed through XML report parsing rules to obtain corresponding structured medical data.
  • the preset parsing rules include medical image text recognition rules, DICOM file parsing rules, and XML report parsing rules, while medical image text recognition rules, DICOM file parsing rules, and XML report parsing rules cover a wide range, so The computer equipment can comprehensively parse the unstructured medical data to be stored, and the usage scenarios of the preset parsing rules are expanded.
  • the computer equipment can acquire the unstructured medical data to be stored through passive reception, or can acquire the unstructured medical data to be stored through active acquisition.
  • the specific implementation methods of these two methods are described in detail below:
  • the first type if the computer equipment is the unstructured medical data to be stored acquired through passive reception, the above step 2310 includes: receiving the unstructured medical data to be stored sent by the corresponding medical equipment through a preset application program interface. Structured medical data.
  • the computer device receives the unstructured medical data to be stored sent by the corresponding medical device through a preset application program interface.
  • the computer device can receive the unstructured medical data to be stored sent by the CT device through an application program interface with the CT device.
  • the computer device can completely receive the unstructured medical data to be stored sent by the corresponding medical device through the preset application program interface, thereby ensuring that the unstructured medical data to be stored received by the computer device is data integrity.
  • the second type if the computer equipment is the unstructured medical data to be stored that is acquired through active acquisition, the above step 2310 includes: searching in the medical equipment corresponding to the target address identifier according to the target address identifier, and obtaining the unstructured medical data to be stored. Stored unstructured medical data.
  • the computer equipment searches the medical equipment corresponding to the target address identifier according to the target address identifier, and obtains the unstructured medical data to be stored.
  • the target address identifier of the MR device can correspond to the target address identifier of the MR device according to the target address identifier of the MR device. Search in the MR equipment to obtain the unstructured medical data to be stored.
  • the computer equipment can quickly search the medical equipment corresponding to the target address identifier according to the target address identifier, so that the unstructured medical data to be stored can be quickly obtained, which improves the ability to obtain the unstructured medical data to be stored. Efficiency of medical data.
  • the above method further includes:
  • Step 2510 Receive a query request; the query request carries the data identifier of the data to be queried.
  • the computer device receives a query request, wherein the query request carries a data identifier of the data to be queried.
  • the data identifier of the data to be queried may be the data identifier of unstructured medical data or the data identifier of structured medical data.
  • the query request received by the computer device may be one query request or multiple query requests.
  • Step 2520 based on the query request, according to the data identifier and the corresponding relationship, perform inquiries in the fourth database and the fifth database, respectively, to obtain structured medical data and unstructured medical data corresponding to the data identifier.
  • the computer equipment respectively in the above fourth The database and the fifth database are queried to obtain structured medical data and unstructured medical data corresponding to the data identifier of the data to be queried carried in the query request.
  • the computer device may perform a query in the above-mentioned fourth database based on the data identifier of the data to be queried carried in the query request based on the received query request, and obtain the corresponding data identifier of the data to be queried carried in the query request.
  • the computer device can, according to the data identifier of the data to be queried carried in the query request, and the correspondence between the unstructured medical data to be stored and the corresponding structured medical data,
  • the fourth database and the fifth data are respectively queried accurately, and the structured medical data and unstructured medical data corresponding to the data identifier of the data to be queried carried in the query request are accurately obtained, and the structure corresponding to the obtained query request is improved.
  • the structured medical data and unstructured medical data corresponding to the data identifier of the data to be queried carried in the query request can be quickly queried in the fourth database and the fifth database, and the data of the data to be queried carried in the query request can be obtained. Efficiency of identifying corresponding structured medical data and unstructured medical data.
  • the method may include:
  • the preset parsing rules include: medical image text recognition rules, DICOM file parsing rules and XML Report parsing rules.
  • the structured medical data and the corresponding relationship are stored in the preset fourth database, and the unstructured medical data is stored in the preset fifth database.
  • S5 Receive a query request; the query request carries a data identifier of the data to be queried.
  • S6 based on the query request, perform a query in the fourth database according to the data identifier to obtain structured medical data corresponding to the data identifier.
  • an unstructured data storage device 2700 including: an acquisition module 2710, a parsing module 2720 and a storage module 2730, wherein:
  • the obtaining module 2710 is used for obtaining unstructured medical data to be stored.
  • the parsing module 2720 is used for parsing the unstructured medical data to obtain structured medical data corresponding to the unstructured medical data and the corresponding relationship between the unstructured medical data and the structured medical data.
  • the storage module 2730 is configured to store the structured medical data and the corresponding relationship in the preset fourth database, and store the unstructured medical data in the preset fifth database.
  • the unstructured data storage device 2700 provided in this embodiment can execute the foregoing method embodiments, and its implementation principles and technical effects are similar, and details are not described herein again.
  • the above-mentioned parsing module 2720 includes: a parsing unit and a establishing unit, wherein:
  • the parsing unit is used for parsing the unstructured medical data by adopting a preset parsing rule to obtain structured medical data corresponding to the unstructured medical data.
  • the establishment unit is used to establish the corresponding relationship between the unstructured medical data and the structured medical data according to the identification of the unstructured medical data and the identification of the structured medical data.
  • the preset parsing rules include: medical image text recognition rules, DICOM file parsing rules and XML report parsing rules.
  • the unstructured data storage device provided in this embodiment can execute the foregoing method embodiments, and the implementation principles and technical effects thereof are similar, and details are not described herein again.
  • the foregoing obtaining module 2710 includes: a first obtaining unit, wherein:
  • the first acquiring unit is configured to receive, through a preset application program interface, unstructured medical data to be stored sent by a corresponding medical device.
  • the unstructured data storage device provided in this embodiment can execute the above method embodiments, and its implementation principle and technical effect are similar, and details are not repeated here.
  • the foregoing obtaining module 2710 includes: a second obtaining unit, wherein:
  • the second obtaining unit is configured to search in the medical equipment corresponding to the target address identifier according to the target address identifier to obtain unstructured medical data to be stored.
  • the unstructured data storage device provided in this embodiment can execute the foregoing method embodiments, and the implementation principles and technical effects thereof are similar, and details are not described herein again.
  • the foregoing apparatus further includes: a receiving module and a querying module, wherein:
  • the receiving module is used for receiving a query request; the query request carries a data identifier of the data to be queried.
  • the query module is used for querying the fourth database and the fifth database respectively based on the query request, according to the data identification and the corresponding relationship, to obtain structured medical data and unstructured medical data corresponding to the data identification.
  • the unstructured data storage device provided in this embodiment can execute the foregoing method embodiments, and the implementation principles and technical effects thereof are similar, and details are not described herein again.
  • the above query module includes: a first query unit and a second query unit, wherein:
  • the first query unit is configured to perform a query in the fourth database according to the data identifier based on the query request to obtain structured medical data corresponding to the data identifier.
  • the second query unit is configured to perform a query in the fifth database according to the structured medical data corresponding to the data identifier and the corresponding relationship to obtain unstructured medical data corresponding to the data identifier.
  • the unstructured data storage device provided in this embodiment can execute the foregoing method embodiments, and the implementation principles and technical effects thereof are similar, and details are not described herein again.
  • Each module in the above-mentioned unstructured data storage device may be implemented in whole or in part by software, hardware and combinations thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • the present application provides a computer-readable storage medium having stored thereon a computer program that is executed by a processor (eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU 330, or one or more modules of FIG. 27), when executed, implements at least one of the processes 2300, 2400, 2500, 2600-2, and 2600-2.
  • a processor eg, data processing system 110, user terminal 130, processor 210, CPU 340, GPU 330, or one or more modules of FIG. 27
  • Figure 28 is an exemplary flow diagram of a medical system workflow 2800 according to some embodiments of the present application.
  • Process 2800 includes the workflow of performing imaging examinations in medical system 100 .
  • the process 2800 may include the following steps.
  • Step 2810 obtaining an application form.
  • the application form can be a document that records a request for taking medical images.
  • the format of the application form may be a document, a picture, or the like.
  • the request for taking a medical image may be recorded on the application form in the form of text.
  • the request for taking a medical image may include information related to taking a medical image, eg, patient-related information, photographing-related information, and the like. Exemplary patient-related information may include name, age, gender, and the like.
  • Exemplary shot-related information may include regions of interest (eg, brain, chest, lungs, abdomen, heart, bones, etc.), image type (eg, CT images, MR images, PET images, X-rays, ultrasound images, etc.) ), shooting time, etc.
  • regions of interest eg, brain, chest, lungs, abdomen, heart, bones, etc.
  • image type eg, CT images, MR images, PET images, X-rays, ultrasound images, etc.
  • shooting time etc.
  • the requisition may be stored, eg, in a hospital information system (HIS), or in a system associated with image capture, eg, a radiology information system (RIS), an imaging system.
  • the application form is unstructured data, and the application form can be stored according to the unstructured data storage methods shown in Figure 23- Figure 28.
  • the application form is parsed to obtain structured medical data corresponding to the application form, and a corresponding relationship between the application form and the structured medical data.
  • the structured medical data and the corresponding relationship are stored in a fourth database, and the application form is stored in a fifth database.
  • a hospital information system (HIS) and a radiology information system (RIS) or imaging system may communicate over the network 120 .
  • the hospital information system (HIS) obtains the application form, it can send the application form to the radiology information system (RIS) or the imaging system through the network 120, and the radiology information system (RIS) or the imaging system can arrange the shooting after receiving the application form.
  • the Radiology Information System (RIS) or imaging system can obtain relevant information in the application form by extracting and recognizing the text in the paper application form. For example, the text in the application form can be extracted and recognized through a preset recognition algorithm.
  • the preset recognition algorithm may include an optical character recognition algorithm (OCR, Optical Character Recognition).
  • OCR optical character recognition algorithms
  • Optical character recognition algorithms may include but are not limited to CTC (Connectionist Temporal Classification) algorithm, CRNN (Convolutional Recurrent Neural Network) algorithm, CPTN (Connectionist Text Proposal Network) algorithm, mutli-label classification algorithm, etc.
  • error detection and correction may be performed on the text in the extracted application form first.
  • the text in the extracted application form can be used as the text to be processed for error detection, and the type and location of the error text can be obtained.
  • a target error correction strategy corresponding to the type of the error text is determined.
  • Step 2820 performing image inspection to obtain a target image.
  • the imaging examination can be arranged according to the application form to obtain the target image.
  • Step 2830 generate an image report based on the target image.
  • a radiology information system (RIS) or imaging system can generate an image report based on the target image after generating the target image.
  • a radiology information system (RIS) or imaging system can send the generated target images to a PACS and/or post-processing workstation, which can generate image reports from the target images.
  • the image report can be automatically generated according to the image report generation method of FIGS. 4-11 .
  • the method for automatically generating an image report may include the following steps: acquiring a target image; cutting out a first characteristic image from the target image, wherein the first characteristic image includes an image of a region of interest in the target image; obtaining a second feature image matching the first feature image; obtaining a knowledge graph, where the knowledge graph includes the association relationship between the second feature image and the corresponding image description information and diagnosis result information; based on the knowledge graph and the In the second characteristic image, image description information and diagnosis result information corresponding to the second characteristic image are found; and an image report of the target image is generated according to the target image, the image description information and the diagnosis result information.
  • a knowledge graph needs to be constructed in advance.
  • the obtained original text can be processed for error correction first, and then stored.
  • the acquired original text can be used as the text to be processed, and the acquired original text can be corrected for errors according to the text error correction methods shown in FIGS. 12 to 22 .
  • the original text and original images obtained from books, newspapers, websites, electronic medical records, laboratory tests, diagnostic reports, etc., as well as characteristic images can be unstructured data, and can be stored according to the unstructured data storage methods shown in Figures 23-27.
  • Original text, original image, and feature image are stored.
  • Step 2840 performing error detection and correction on the image report.
  • error detection can also be performed on the text in the image report to obtain the type and location of the error text; according to the correspondence between the error text type and the error correction strategy, the target correction corresponding to the type of error text is determined. Error strategy; according to the target error correction strategy and the position of the error text, the error text is corrected and a formal report is generated. Further description can be found elsewhere in this specification, eg, Figures 12-22.
  • Step 2850 store the image report.
  • the formal report may also be parsed to obtain structured medical data corresponding to the formal report, and the correspondence between the formal report and the structured medical data;
  • the structured medical data and the corresponding relationship are stored in the preset fourth database, and the formal report is stored in the preset fifth database.
  • Step 2860 retrieve the image report.
  • a user may input a query request through the user terminal 130, and the data processing system 110 may search the fourth database for structured data corresponding to the input query request, and then use the unstructured data to search for structured data corresponding to the input query request.
  • Correspondence to Structured Data Corresponding unstructured data (eg, image reports) are looked up from the fifth database. More descriptions on obtaining, storing, and retrieving unstructured data can be found elsewhere in this specification, eg, Figures 23-27.
  • a program portion of the technology may be considered a "product” or “article of manufacture” in the form of executable code and/or related data, which is involved or implemented through a computer-readable medium.
  • Tangible, permanent storage media includes memory or storage used by any computer, processor, or similar device or associated module. For example, various semiconductor memories, tape drives, magnetic disk drives, or similar devices that can provide storage functions for software at any time.
  • All or part of the software may sometimes communicate over a network, such as the Internet or other communication network.
  • a network such as the Internet or other communication network.
  • Such communications enable the loading of software from one computer device or processor to another.
  • a hardware platform loaded from a management server or host computer of an imaging system to a computer environment, or other computer environment implementing the system, or a system of similar functionality related to providing information needed for on-demand services.
  • another medium capable of transmitting software elements can also be used as a physical connection between local devices, such as light waves, radio waves, electromagnetic waves, etc., through cables, optical cables or air.
  • the physical medium used for the carrier wave such as a cable, wireless connection, or fiber optic cable, etc., can also be considered to be the medium that carries the software.
  • Unless the usage herein is limited to tangible "storage” media, other terms referring to computer or machine "readable media” refer to media that participate in the execution of any instructions by a processor.
  • a computer-readable medium may take many forms, including, but not limited to, tangible storage media, carrier wave media, or physical transmission media.
  • Stable storage media include optical or magnetic disks, and other storage systems used in computers or similar devices that can implement the system components described in the figures.
  • Unstable storage media include dynamic memory, such as the main memory of a computer platform.
  • Tangible transmission media include coaxial cables, copper cables, and fiber optics, including the wires that form a bus within a computer system.
  • Carrier-wave transmission media can transmit electrical, electromagnetic, acoustic, or light wave signals generated by radio frequency or infrared data communications methods.
  • Typical computer readable media include hard disks, floppy disks, magnetic tapes, any other magnetic media; CD-ROMs, DVDs, DVD-ROMs, any other optical media; punch cards, any other physical storage media containing a pattern of pinholes; RAM, PROM , EPROM, FLASH-EPROM, any other memory chips or magnetic tapes; carrier waves, cables or connecting means for transmitting carrier waves, any other program code and/or data that can be read by a computer. Many of these computer-readable media may be present in the process of executing instructions by a processor, delivering one or more results.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Library & Information Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种方法，可以包括获取目标影像；从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；获取与所述第一特征影像匹配的第二特征影像；获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。

Description

医疗数据处理方法与系统
交叉引用
本申请要求以下申请的优先权：
2020年7月17日提交的编号为202010690317.8的中国申请；
2020年12月10日提交的编号为202011434314.4的中国申请；
2020年12月17日提交的编号为202011495360.5的中国申请；
上述申请的内容以引用方式被包含于此。
技术领域
本说明书涉及医疗领域，特别涉及一种医疗数据处理方法和系统。
背景技术
随着医疗领域的发展，对医疗数据的处理越来越重要，例如，医学影响报告的生成，医疗文本的纠错，医疗数据的存储与检索等。因此，需要提供一种医疗数据处理方法和系统，用于高效、准确地处理医疗数据。
简述
本申请一方面提供了一种方法，实现于包括至少一个存储设备和至少一个处理设备的装置上。所述方法可以包括：获取目标影像；从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；获取与所述第一特征影像匹配的第二特征影像；获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
在一些实施例中，构建所述知识图谱包括：确定所述知识图谱的图谱结构；获取预设领域的原始文本和原始影像；从所述原始文本和所述原始影像中提取出多个实体数据；确定所述多个实体数据之间的关联关系；根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述领域知识图谱。
在一些实施例中，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：将所述原始文本输入到预先训练的分词模型中，得到所述分词模型输出的多个分词结果；将所述多个分词结果输入到预先训练的实体识别模型中，得到所述实体识别模型输出的特征文本，并将所述特征文本确定为所述实体数据。
在一些实施例中，所述确定所述多个实体数据之间的关联关系，包括：将所述原始文本和多个所述特征文本输入到预先训练的关系提取模型中，得到所述关系提取模型输出的多个所述特征文本之间的关联关系。
在一些实施例中，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：将所述原始影像输入到预先训练的识别模型中，得到所述识别模型输出的识别结果，所述识别结果用于指示所述原始影像中的感兴趣区域；从所述原始影像中截取出所述感兴趣区域的特征影像；将所述感兴趣区域的所述特征影像确定为所述实体数据。
在一些实施例中，所述从所述原始文本和所述原始影像中提取出多个实体数据，确定所述多个实体数据之间的关联关系，包括：从所述原始文本中提取出所述原始影像对应的影像描述文本和诊断结果文本；从所述原始影像中提取出特征影像；将所述影像描述文本，所述诊断结果文本，和所述原始影像的所述特征影像确定为所述实体数据；建立所述原始影像的所述特征影像、所述影像描述文本和所述诊断结果文本之间的关联关系。
在一些实施例中，所述根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述知识图谱，包括：根据所述多个实体数据之间的关联关系将所述多个实体数据组成多个三元组；将所述多个三元组填充到所述图谱结构中，得到所述知识图谱。
在一些实施例中，所述方法还可以包括：基于所述领域知识图谱和所述第二特征影像，查找与所述第二特征影像对应的疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个；根据所述目标影像、所述疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个，生成所述目标影像的影像报告。
在一些实施例中，所述方法还可以包括：对所述目标影像的影像报告进行错误检测，得到错误文本的类型和位置；根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
在一些实施例中，所述错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本；根据所述至少一个候选文本，对所述错误文本进行纠错处理。
在一些实施例中，若所述错误文本的类型包括错字类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：根据所 述错字类型的文本的位置在所述影像报告中提取出所述错字类型的文本；从第一数据库中选取出与所述错字类型的文本同音，和/或，与所述错字类型的文本之间的相似度大于相似度阈值的文本作为所述候选文本。
在一些实施例中，若所述错误文本的类型包括缺失类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：根据所述缺失类型的文本的位置在所述影像报告中提取出与所述缺失类型的文本相邻的文本；从第二数据库中选取出与所述相邻的文本匹配的文本作为所述候选文本。
在一些实施例中，若所述错误文本的类型包括无序类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：根据所述无序类型的文本的位置在所述影像报告中提取出所述无序类型的文本；从第三数据库中选取出与所述无序类型的文本包含相同字的文本作为所述候选文本。
在一些实施例中，所述根据所述至少一个候选文本，对所述错误文本进行纠错处理，包括：将至少一个所述候选文本依次添加到所述错误文本对应位置上，得到待处理文本；对所述待处理文本的通顺程度进行打分；按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
在一些实施例中，所述错误文本的类型包括冗余类型，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本；若确定删除，则删除所述错误文本；若确定不删除，则不执行删除操作。
在一些实施例中，所述根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本，包括：根据所述冗余类型的文本的位置在所述待影像报告中提取出包含所述冗余类型的文本；检测包含所述冗余类型的文本是否通顺；若不通顺，则确定删除所述冗余类型的文本；若通顺，则确定不删除所述冗余类型的文本。
在一些实施例中，所述影像报告为非结构化医疗数据，所述方法进一步包括：对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系；将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述影像报告存储到第五数据库中。
在一些实施例中，所述对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系，包括：根据所述影像报告的非结构化医疗数据类型，从多个预设的解析规则中查找对应得解析规则；根据所述对 应解析规则，对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据；根据所述影像报告的非结构化医疗数据的标识和所述结构化医疗数据的标识，建立所述影像报告和所述结构化医疗数据之间的对应关系。
在一些实施例中，所述多个预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
在一些实施例中，所述方法进一步包括：通过预设的应用程序接口，接收对应的医疗设备发送的所述影像报告；或者，根据目标地址标识，在所述目标地址标识对应的医疗设备中进行查找，得到所述影像报告。
在一些实施例中，所述方法还包括：接收查询请求，所述查询请求携带有所述影像报告的数据标识；基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告。
在一些实施例中，所述基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告，包括：基于所述查询请求，根据所述数据标识在所述第四数据库中进行查询，得到所述数据标识对应的结构化医疗数据；根据所述数据标识对应的结构化医疗数据和所述对应关系，在所述第五数据库中进行查询，得到所述影像报告。
本申请另一方面提供了一种系统，所述系统包括至少一个存储设备和至少一个处理器。所述至少一个存储设备包括一组指令。所述至少一个处理器与所述至少一个存储设备通信，其中当所述至少一个处理器执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：获取目标影像；从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；获取与所述第一特征影像匹配的第二特征影像；获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
本申请另一方面提供了一种系统，包括：影像截取模块，被配置为获取目标影像；以及从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像。所述系统还可以包括影像查找模块，被配置为获取与所述第一特征影像匹配的第二特征影像。所述系统还可以包括信息查找模块，被配置为获取知识图谱，所述知 识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；以及基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息。所述系统还可以包括报告生成模块，被配置为根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
本申请另一方面提供了一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：获取目标影像；从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；获取与所述第一特征影像匹配的第二特征影像；获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
本申请另一方面提供了一种数据处理方法，实现于包括至少一个存储设备和至少一个处理设备的装置上。所述方法包括：对待处理文本进行错误检测，得到错误文本的类型和位置；根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
本申请另一方面提供了一种系统，所述系统包括至少一个存储设备和至少一个处理器。所述至少一个存储设备包括一组指令。所述至少一个处理器与所述至少一个存储设备通信，其中当所述至少一个处理器执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：对待处理文本进行错误检测，得到错误文本的类型和位置；根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
本申请另一方面提供了一种系统，包括：检测模块，被配置为对待处理文本进行错误检测，得到错误文本的类型和位置；确定模块，被配置为根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；纠错模块，被配置为根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
本申请另一方面提供了一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：对待处理文本进行错误检测，得到错误文本的类型 和位置；根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
本申请另一方面提供了一种非结构化数据存储方法，实现于包括至少一个存储设备和至少一个处理设备的装置上，所述方法包括：获取待存储的非结构化医疗数据；对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
本申请另一方面提供了一种系统，所述系统包括至少一个存储设备和至少一个处理器。所述至少一个存储设备包括一组指令。所述至少一个处理器与所述至少一个存储设备通信，其中当所述至少一个处理器执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：获取待存储的非结构化医疗数据；对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
本申请另一方面提供了一种系统，包括：获取模块，被配置为获取待存储的非结构化医疗数据；解析模块，被配置为对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；存储模块，被配置为将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
本申请另一方面提供了一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：获取待存储的非结构化医疗数据；对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
附图描述
在此所述的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请并不构成对本申请的限定。在各图中，相同标号表示相同部件。
图1是根据本申请一些实施例所示的医疗系统的应用场景示意图；
图2是根据本申请的一些实施例所示的计算设备的示例性硬件和/或软件的示意图；
图3是根据本申请的一些实施例所示的一种移动设备的示例性硬件和/或软件示意图；
图4是根据本申请一些实施例所示的一种影像报告生成方法的示例性流程图；
图5是根据本申请一些实施例所示的构建领域知识图谱步骤的示例性流程图；
图6是根据本申请一些实施例所示的图谱结构的示意图；
图7是根据本申请一些实施例所示的领域知识图谱的示意图；
图8是根据本申请一些实施例所示的查找与目标影像对应的影像描述信息和诊断结果信息的示例性流程图；
图9是根据本申请一些实施例所示的生成影像报告的示例性流程图；
图10是根据本申请一些实施例所示的影像报告的示意图；
图11是根据本申请一些实施例所示的影像报告生成装置的示例性框图；
图12A是根据本申请一些实施例所示的文本纠错方法的示例性流程图；
图12B是根据本申请一些实施例所示的训练方法的示例性流程图；
图13是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理的示例性流程图；
图14是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本的示例性流程图；
图15是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本的示例性流程图；
图16是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本的示例性流程图；
图17是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，对错误文本进行纠错的示例性流程图；
图18是根据本申请一些实施例所示的根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理的示例性流程图；
图19是根据本申请一些实施例所示的冗余类型错误文本纠错的示例性流程图；
图20是根据本申请一些实施例所示的数据处理方法的是示例性流程图；
图21是根据本申请一些实施例所示的纠错系统结构的示意图；
图22是根据本申请一些实施例所示的数据处理装置结构的框图；
图23是根据本申请一些实施例所示的非结构化数据存储方法的示例性流程图；
图24是根据本申请一些实施例所示的非结构化数据存储方法的示例性流程图；
图25是根据本申请一些实施例所示的非结构化数据存储方法的示例性流程图；
图26A是根据本申请一些实施例所示的非结构化数据存储方法的示例性流程图；
图26B是根据本申请一些实施例所示的非结构化数据存储方法的示例性流程图；
图27是根据本申请一些实施例所示的非结构化数据存储装置的结构框图；以及
图28是根据本申请一些实施例所示的医疗系统工作流的示例性流程图。
具体描述
为了更清楚地说明本申请的实施例的技术方案，下面将对实施例描述中所需要使用的附图作简单的介绍。显而易见地，下面描述中的附图仅仅是本申请的一些示例或实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图将本申请应用于其他类似情景。应当理解，给出这些示例性实施例仅仅是为了使相关领域的技术人员能够更好地理解进而实现本发明，而并非以任何方式限制本发明的范围。除非从语言环境中显而易见或另做说明，图中相同标号代表相同结构或操作。
如本申请和权利要求书中所示，除非上下文明确提示例外情形，“一”、“一个”、“一种”和/或“该”等词并非特指单数，也可包括复数。一般说来，术语“包括”与“包含”仅提示包括已明确标识的步骤和元素，而这些步骤和元素不构成一个排它性的罗列，方法或者设备也可能包含其他的步骤或元素。
虽然本申请对根据本申请的实施例的系统中的某些模块做出了各种引用，然而，任何数量的不同模块可以被使用并运行在客户端和/或服务器上。所述模块仅是说明性的，并且所述系统和方法的不同方面可以使用不同模块。
本申请中使用了流程图用来说明根据本申请的实施例的系统所执行的操作。应当理解的是，前面或下面操作不一定按照顺序来精确地执行。相反，可以按照倒序或同时处理各种步骤。同时，也可以将其他操作添加到这些过程中，或从这些过程移除某一步或数步操作。
本申请一方面提供了影像报告生成方法、装置、计算机设备和存储介质，获取 目标影像，并从目标影像中截取出第一特征影像；根据第一特征影像在预先设置的特征影像库中进行查找，得到与第一特征影像匹配的第二特征影像；基于预先构建的领域知识图谱和第二特征影像，查找出与目标影像对应的影像描述信息和诊断结果信息；根据目标影像、影像描述信息和诊断结果信息生成影像报告。本公开实施例中，终端可以利用预先构建的领域知识图谱自动生成影像报告，与现有技术相比，无需阅片医生手动填写影像报告，提高了影像报告的生成效率。
本申请另一方面提供了数据处理方法、装置、计算机设备和存储介质，通过对待处理文本进行错误检测，得到错误文本的类型和位置，并根据错误文本类型和纠错策略之间的对应关系，确定与错误文本的类型对应的目标纠错策略，进一步的根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理。在上述纠错的数据处理方法中，由于每种类型的错误文本对应各自的目标纠错策略，实现了针对每种错误文本类型的纠错操作，相比于传统的无论任何类型的错误文本均采用大数据库获取用于纠错的文本导致的数据处理量大和纠错精度差的问题，本申请所述的纠错方法极大的提高了纠错的准确性，以及提高了数据处理速度。
本申请另一方面提供了非结构化数据存储方法、装置、计算机设备和存储介质，通过对待存储的非结构化医疗数据进行解析，能够得到该非结构化医疗数据对应的结构化医疗数据，以及该非结构化医疗数据与得到的结构化医疗数据之间的对应关系，从而可以将得到的结构化医疗数据和得到的对应关系存储到预设的第一数据库中，将该待存储的非结构化医疗数据存储到预设的第二数据库中，由于对非结构化医疗数据进行解析的过程十分简单，能够快速准确地得到待存储的非结构化医疗数据对应的结构化医疗数据，以及该非结构化医疗数据与得到的结构化医疗数据之间的对应关系，从而能够快速地将得到的结构化医疗数据和得到的对应关系存储到预设的第一数据库中，将待存储的非结构化医疗数据存储到预设的第二数据库中，从而提高了对待存储的非结构化医疗数据的存储效率。
图1是根据本申请一些实施例所示的医疗系统100的应用场景示意图。在一些实施例中，医疗系统100可以包括数据处理系统110、网络120、用户终端130以及存储设备140。
数据处理系统110可以用于处理医疗系统100的医疗数据。数据处理系统110可以包括图像存档与传输系统(PACS)，后处理工作站，手术系统，医院信息系统(HIS)，实验室信息系统(LIS)，放射科信息系统(RIS)，医生工作站，护士工作站， 成像系统，住院管理系统，门急诊管理系统，收费系统等中的一种或几种的组合。
在一些实施例中，数据处理系统110可以包括一个或多个服务器。服务器可以用于管理资源以及处理来自本系统至少一个组件或外部系统的数据和/或信息。在一些实施例中，服务器可以是独立的服务器或者服务器组。该服务器组可以是集中式的或者分布式的(如：服务器可以是分布系统)。在一些实施例中该服务器可以是本地的或者远程的。例如，服务器可通过网络120访问存储于存储设备140中的信息和/或资料。在一些实施例中，服务器可在云平台上执行。例如，该云平台可包括私有云、公共云、混合云、社区云、分散式云、内部云等中的一种或其任意组合。
在一些实施例中，服务器可以包括处理设备。在一些实施例中，处理设备可包含一个或多个子处理设备(例如，单芯处理设备或多核多芯处理设备)。仅仅作为范例，处理设备可包含中央处理器(CPU)、专用集成电路(ASIC)、专用指令处理器(ASIP)、图形处理器(GPU)、物理处理器(PPU)、数字信号处理器(DSP)、现场可编程门阵列(FPGA)、可编辑逻辑电路(PLD)、控制器、微控制器单元、精简指令集电脑(RISC)、微处理器等或以上任意组合。
医疗系统100之间各组成部件之间可以通过网络120进行数据和/或信息的交换。在一些实施例中，医疗系统100中的一个或多个组件(例如，数据处理系统110，用户终端130，存储设备140)可通过网络120发送数据和/或信息给医疗系统100中的其他组件。在一些实施例中，网络120可是任意类型的有线或无线网络。例如，网络120可包括缆线网络、有线网络、光纤网络、电信网络、内部网络、网际网络、区域网络(LAN)、广域网络(WAN)、无线区域网络(WLAN)、都会区域网络(MAN)、公共电话交换网络(PSTN)、蓝牙网络、ZigBee网络、近场通讯(NFC)网络等或以上任意组合。在一些实施例中，网络120可包括一个或多个网络进出点。例如，网络120可包含有线或无线网络进出点，如基站和/或网际网络交换点，通过这些进出点，医疗系统100的一个或多个组件可连接到网络120上以交换数据和/或信息。
用户终端130可以与数据处理系统110进行数据交互。在一些实施例中，用户终端130可以从存储设备140和/或数据处理系统110接收数据和/或信息。在一些实施例中，用户终端130可以包括手机(141-1、141-2、141-3...)、平板电脑(142-1、142-2、142-3...)、笔记本电脑(143-1、143-2、143-3...)等中的一种或其任意组合。在一些实施例中，用户终端130可以通过用户终端130中的应用软件与数据处理系统110建立通信。例如，用户终端130可以通过应用软件接收、显示来自存储设备140和/或数据 处理系统110的数据和/或信息。又例如，用户终端130可以通过应用软件向存储设备140和/或数据处理系统110发送指令和/或请求。
医疗系统100的一个或多个组件可通过网络120访问存储于存储设备140中的数据或指令。数据指对信息的数字化表示，可以包括各种类型，比如二进制数据、文本数据、图像数据、视频数据等。指令指可控制设备或器件执行特定功能的程序。存储设备140可与网络120连接，以与医疗系统100的一个或多个组件(例如，数据处理系统110，用户终端130)通讯。在一些实施例中，存储设备140可以是服务器的一部分。存储设备140可以在单个中央服务器、通过通信链路连接的多个服务器或多个个人设备中实现。存储设备140可以由多个个人设备和云服务器生成。在一些实施例中，存储设备140可包括大容量存储器、可移动存储器、挥发性读写存储器(例如，随机存取存储器RAM)、只读存储器(ROM)等或以上任意组合。在一些实施例中，数据库可在云平台上实现。例如，该云平台可包括私有云、公共云、混合云、社区云、分散式云、内部云等或以上任意组合。
图2是根据本申请的一些实施例所示的计算设备的示例性硬件和/或软件的示意图。如图2所示，计算设备200可以包括处理器210、存储器220、输入/输出(I/O)接口230和通信端口240。在一些实施例中，数据处理系统110的服务器可以根据处理设备200实现。
处理器210可以执行计算指令(程序代码)并执行本申请描述的医疗系统100的功能。所述计算指令可以包括程序、对象、组件、数据结构、过程、模块和功能(所述功能指本申请中描述的特定功能)。例如，处理器210可以处理从医疗系统100的任何组件获得的目标影像。在一些实施例中，处理器210可以包括微控制器、微处理器、精简指令集计算机(RISC)、专用集成电路(ASIC)、应用特定指令集处理器(ASIP)、中央处理器(CPU)、图形处理单元(GPU)、物理处理单元(PPU)、微控制器单元、数字信号处理器(DSP)、现场可编程门阵列(FPGA)、高级RISC机(ARM)、可编程逻辑器件以及能够执行一个或多个功能的任何电路和处理器等，或其任意组合。仅为了说明，图2中的计算设备200只描述了一个处理器，但需要注意的是本申请中的计算设备200还可以包括多个处理器。
存储器220可以存储从医疗系统100的任何其他组件获得的数据/信息。在一些实施例中，存储器220可以包括大容量存储器、可移动存储器、易失性读取和写入存储器和只读存储器(ROM)等，或其任意组合。示例性大容量存储器可以包括磁盘、光盘和固态驱动器等。可移动存储器可以包括闪存驱动器、软盘、光盘、存储卡、压缩盘和磁带等。易失性读取和写入存储器可以包括随机存取存储器(RAM)。RAM可以包括动态RAM (DRAM)、双倍速率同步动态RAM(DDR SDRAM)、静态RAM(SRAM)、晶闸管RAM(T-RAM)和零电容(Z-RAM)等。ROM可以包括掩模ROM(MROM)、可编程ROM(PROM)、可擦除可编程ROM(PEROM)、电可擦除可编程ROM(EEPROM)、光盘ROM(CD-ROM)和数字通用盘ROM等。
输入/输出接口(I/O)230可以用于输入或输出信号、数据或信息。在一些实施例中，输入/输出接口230可以使用户与医疗系统100进行联系。在一些实施例中，输入/输出接口(I/O)230可以包括输入装置和输出装置。示例性输入装置可以包括键盘、鼠标、触摸屏和麦克风等中的一种或以上任意组合。示例性输出设备可以包括显示设备、扬声器、打印机、投影仪等，或其任意组合。示例性显示装置可以包括液晶显示器(LCD)、基于发光二极管(LED)的显示器、平板显示器、曲面显示器、电视设备、阴极射线管(CRT)等中的一种或以上任意组合。通信端口240可以连接到网络以便数据通信。所述连接可以是有线连接、无线连接或两者的组合。有线连接可以包括电缆、光缆或电话线等，或其任意组合。无线连接可以包括蓝牙、Wi-Fi、WiMax、WLAN、ZigBee、移动网络(例如，3G、4G或5G等)等中的一种或以上任意组合。在一些实施例中，通信端口240可以是标准化端口，如RS232、RS485等。在一些实施例中，通信端口240可以是专门设计的端口。例如，通信端口240可以根据数字成像和医学通信协议(DICOM)进行设计。
根据本申请的一些实施例，图3是根据本申请的一些实施例所示的一种移动设备的结构示意图，该移动设备能够被配置为实现实施本申请中披露的特定系统，例如，用户终端130。在本例中，用于显示和交互位置相关信息的用户设备是一个移动设备300，可以包括但不限于，智能手机、平板电脑、音乐播放器、便携游戏机、全球定位系统(Global Positioning System,GPS)接收器、可穿戴计算设备(如眼镜、手表等)，或者其他形式。本例中的移动设备300可以包括一个或多个中央处理器(CPUs)340、图形处理器(Graphical Processing Units,GPUs)330、显示单元320、内存360、天线310(例如，无线通信单元)、存储单元370，以及一个或多个输入/输出(input output(I/O))单元350。任何其他合适的组件，可以包括但不限于系统总线或控制器(图上未显示)，也可能被包括在移动设备300中。如图3所示，一个移动操作系统361，如iOS、Android、Windows Phone等，以及一个或多个应用362可以从存储单元370加载进内存360中，并被中央处理器340所执行。应用362可以包括一个浏览器或其他适合在移动设备300上接收并处理影像报告的移动应用。用户(例如，医生)关于影像报告的交互可以通过输入/输出(input output(I/O))单元350获得并提供给服务器120，以及/或医疗系统100的其他组件，例如：通过网络120。
为了实现不同的模块、单元以及在之前的披露中所描述的他们的功能，计算机硬件平台可以被用作以上描述的一个或多个元素的硬件平台(例如：服务器，以及/或医疗系统100的其他组件)。这类计算机的硬件元素、操作系统和程序语言在自然界中是常见的，可以假定本领域技术人员对这些技术都足够熟悉，能够利用这里描述的技术提供按需服务所需要的信息。一台包含用户界面元素的计算机能够被用作个人计算机(personal computer(PC))或其他类型的工作站或终端设备，被适当程序化后也可以作为服务器使用。可以认为本领域技术人员对这样的结构、程序以及这类计算机设备的一般操作都是熟悉的，因此所有附图也都不需要额外的解释。
随着医疗影像设备的发展，医疗影像在临床诊断中的作用越来越重要。通常情况下，获得医疗影像后都会依据医疗影像形成影像报告。现有的影像报告生成方式是将获得的医疗影像填充到预先提供的影像报告的模板中，模板中未填充的部分由阅片医生手动填写。但是，这种影像报告的生成方式效率很低。本申请提供一种自动生成影像报告的方法和系统。
图4是根据本申请一些实施例所示的一种影像报告生成方法的示例性流程图。在一些实施例中，流程400可以由医疗系统100(如数据处理系统110)执行。例如，流程400可以由PACS，成像系统，或后处理工作站的服务器执行。又例如，流程400可以由处理器210执行。又例如，流程400可以由图11所示的一个或多个模块执行。在一些实施例中，流程400可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程400。下面呈现的流程400的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图4中示出的和下面描述的流程400的操作的顺序不旨在是限制性的。
步骤410，数据处理系统110(例如，影像截取模块1110)可以获取目标影像，并从目标影像中截取出第一特征影像。
在一些实施例中，目标影像可以包括成像系统生成的医学影像，例如，CT(Computed Tomography，电子计算机断层扫描)影像、MR(Magnetic Resonance，磁共振)影像、PET(Positron Emission Computed Tomography，正电子发射型计算机断层显像)影像、X光片(X-Ray)、超声影像等中的至少一种。在本申请中，特征影像可以指感兴趣区域的影像。所述感兴趣区域可以是检测对象(例如，人体，动物，物体等)的整体或其中的一部分，例如，头部、胸部、腹部、心脏、肝脏、上肢、下肢、脊椎、骨骼、血管、病变部位、肿瘤部位等，或者上述部位的任意组合。第一特征影像可以为目标影像的一部分。例 如，目标影像可以为检测对象腹部某一横断面CT图像，第一特征影像可以为腹部CT图像中肝脏的图像。
在一些实施例中，获取目标影像的方式可以有多种，比如通过网络120从影像设备获取目标影像，或者通过网络120从影像数据库(例如，存储设备140，存储器220，PACS等)获取目标影像。本申请实施例对目标影像的获取方式不做限定。
在一些实施例中，获取到目标影像后，可以识别出目标影像中的感兴趣区域，然后将感兴趣区域内的影像截取出来作为第一特征影像。
在一些实施例中，一种确定感兴趣区域的方式，可以包括：将目标影像输入到预先训练的识别模型中，得到识别模型输出的目标影像中的感兴趣区域。另一种确定感兴趣区域的方式，可以包括：用户在用户终端130在目标影像中手动选取感兴趣区域，数据处理系统110接收来自用户终端130的用户针对目标影像的选取操作，根据选取操作确定目标影像中的感兴趣区域，从而获取第一特征影像。本申请实施例对感兴趣区域的确定方式不做限定。
步骤420，数据处理系统110(例如，影像查找模块1120)可以根据第一特征影像在预先设置的特征影像库中进行查找，得到与第一特征影像匹配的第二特征影像。
在一些实施例中，特征影像库(例如，PACS，存储设备140，存储器220等)中可以包括多个特征影像，例如，CT(Computed Tomography，电子计算机断层扫描)影像、MR(Magnetic Resonance，磁共振)影像、PET(Positron Emission Computed Tomography，正电子发射型计算机断层显像)影像、X光片(X-Ray)、超声影像等中的至少一种。
在一些实施例中，预先设置特征影像库，在从目标影像中截取出第一特征影像后，数据处理系统110可以将第一特征影像分别与特征影像库中的各特征影像进行相似度计算，得到多个相似度。数据处理系统110可以将特征影像库中相似度满足预设条件的特征影像确定为第二特征影像。其中，预设条件可以包括相似度大于预设阈值(例如，30％，40％，50％，60％，70％，80％，90％，95％等)。
例如，计算第一特征影像A与特征影像库中的特征影像1之间的相似度1(例如，96％)，计算第一特征影像A与特征影像库中的特征影像2之间的相似度2(例如，70％)，以此类推计算出多个相似度。假如预设阈值设置为95％，则可以将影像特征库中相似度大于95％的特征影像确定第二特征影像，例如，特征影像1。
在一些实施例中，可以通过影像类型(例如，CT影像、MR影像、PET影像、X光片、超声影像等)，感兴趣区域类型(例如，头部、胸部、腹部、心脏、肝脏、上肢、下肢、脊椎、骨骼、血管等)，感兴趣区域形态特征，像素特征(例如，灰度值)等中的至少 一种确定特征影像库中的特征影像和第一特征影像之间的相似度。在一些实施例中，可以通过相似度算法确定特征影像库中的特征影像和第一特征影像之间的相似度，示例性的相似度算法可以包括均值哈希算法、直方图算法等。在一些实施例中，可以通过将特征影像库中的特征影像和第一特征影像输入到训练好的相似度确定模型中，确定特征影像库中的特征影像和第一特征影像之间的相似度。
步骤430，数据处理系统110(例如，信息查找模块1130)可以基于预先构建的领域知识图谱(也可以称为“知识图谱”)和第二特征影像，查找出与第二特征影像对应的影像描述信息和诊断结果信息。
影像描述信息可以为用于描述感兴趣区域特征的信息。在一些实施例中，影像描述信息包括影像中感兴趣区域的形状、尺寸和纹理中的至少一种。诊断结果信息可以为感兴趣区域的与疾病诊断相关的信息。在一些实施例中，诊断结果信息可以包括影像中感兴趣区域的诊断结果。
在一些实施例中，数据处理系统110可以基于预先构建的领域知识图谱和第二特征影像，进一步查找出与第二特征影像对应的疾病、诱因、症状、药物、治疗方案等中的至少一种文本。
在一些实施例中，领域知识图谱可以包括影像和文本的对应关系。例如，领域知识图谱中可以包括影像实体之间的对应关系，文本实体之间的对应关系，和/或影像实体和文本实体之间的对应关系。影像实体可以包括原始影像和/或特征影像。文本实体可以包括影像描述、诊断结果、疾病、诱因、症状、药物、治疗方案等中的至少一种。又例如，领域知识图谱中可以包括历史影像报告和该历史影像报告中的影像之间的对应关系。
在一些实施例中，可以预先构建领域知识图谱，在从特征影像库中查找到与第一特征影像相似的第二特征影像之后，数据处理系统110可以根据第二特征影像在领域知识图谱中进行查找，查找出与第二特征影像关联的影像描述信息和诊断结果信息，将这些影像描述信息和诊断结果信息作为目标影像对应的影像描述信息和诊断结果信息。关于知识图谱的进一步描述可以参见图5-图7。
步骤440，数据处理系统110(例如，报告生成模块1140)可以根据目标影像、影像描述信息和诊断结果信息生成影像报告。
在一些实施例中，影像报告可以包括就诊对象的基础信息(例如，姓名、年龄、性别、检查项目等)、目标影像、影像描述信息和诊断结果信息等。在一些实施例中，可以预先设置影像报告模板，在查找出目标影像的影像描述信息和诊断结果信息之后，数据处理系 统110可以将就诊对象的基础信息、目标影像、影像描述信息和诊断结果信息填充到影像报告模板中，然后自动生成影像报告。
在一些实施例中，影像报告还可以包括疾病、诱因、症状、药物、治疗方案等中的至少一种。数据处理系统110可以将疾病、诱因、症状、药物、治疗方案等中的至少一种填充到影像报告模板中，然后自动生成影像报告。
在一些实施例中，影像报告模板可以存储在医疗系统100的存储设备(例如，存储设备140，存储器220等)中。数据处理系统110可以访问储在医疗系统100的存储设备获取影像报告模板。
在一些实施例中，可以根据图12-图22的纠错方法对生成的目标影像的影像报告进行检错和纠错。在一些实施例中，生成的目标影像的影像报告为非结构化数据，可以根据图23-图27的非结构化数据存储方法对生成的目标影像的影像报告进行存储。
上述影像报告生成方法中，数据处理系统110获取目标影像，并从目标影像中截取出第一特征影像；根据第一特征影像在预先设置的特征影像库中进行查找，得到与第一特征影像匹配的第二特征影像；基于预先构建的领域知识图谱和第二特征影像，查找出与目标影像对应的影像描述信息和诊断结果信息；根据目标影像、影像描述信息和诊断结果信息生成影像报告。本申请实施例中，数据处理系统110可以利用预先构建的领域知识图谱自动生成影像报告，无需阅片医生手动填写影像报告，提高了影像报告的生成效率。
应当注意的是，上述有关流程400的描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以对流程400进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。例如，流程400可以由用户终端130执行，例如，CPU 340和/或GPU 330执行，和/或图11所示的一个或多个模块执行。
图5是根据本申请一些实施例所示的构建领域知识图谱步骤的示例性流程图。在一些实施例中，流程500可以由医疗系统100(如数据处理系统110)执行。例如，流程500可以由PACS，后处理工作站，医生工作站，或成像系统的服务器执行。又例如，流程500可以由处理器210执行。又例如，流程500可以由图11所示的一个或多个模块执行。在一些实施例中，流程500可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程500。下面呈现的流程500的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图5中示出的和下面描述的流程500的操作的顺序 不旨在是限制性的。在一些实施例中，可以基于流程500构建领域知识图谱。
步骤510，数据处理系统110(例如，信息查找模块1130)可以确定领域知识图谱的图谱结构。
图谱结构可以用于表征实体类型和实体连接关系。在一些实施例中，对于医疗领域知识图谱，如表1所示，实体类型可以包括影像实体和/或文本实体。影像实体可以包括原始影像和/或特征影像。文本实体可以包括影像描述、诊断结果、疾病、诱因、症状、药物、治疗方案等中的至少一种。实体连接关系可以表示实体类型之间的关系，包括观察、相克、并列、选择、使用、特征、结果、有关、并发症、包含、诱发、表现等中的至少一种。
表1
实体类型 属性
原始影像 ID、生成日期、路径
特征影像 ID、是否病灶、路径
诊断结果 ID、名称、描述
影像描述 ID、名称、描述
疾病 ID、中医学名、西医学名、描述、易感人群、传染性
诱因 ID、名称、描述、类型
症状 ID、名称、描述、严重程度、身体部位
药物 ID、中医学名、西医学名、描述
治疗方案 ID、描述、治疗周期
在一些实施例中，确定实体类型和实体连接关系后，可以根据上述实体类型和实体连接关系确定如图6所示的图谱结构。
如图6所示，图谱结构中，实体类型可以包括影像实体和文本实体。影像实体可以包括原始影像和特征影像。文本实体可以包括影像描述、诊断结果、疾病、诱因、症状、药物、治疗方案。实体连接关系可以包括观察、相克、并列、选择、使用、特征、结果、有关、并发症、包含、诱发、表现。原始影像和特征影像之间为“特征”关系。特征影像和影像描述之间为“观察”关系。特征影像与诊断结果之间为“结果”关系。特征影像与疾病之间为“观察”关系。影像描述与疾病之间为“观察”关系。诊断结果与疾病之间为“有关”关系。与相同特征影像对应的多个疾病之间为“并发症”关系。疾病与诱因之间为“诱发”关系。疾病与症状之间为“表现”关系。疾病与药物之间为“使用”关系。疾病与治疗方案之间为“使用”关系。治疗方案与药物之间为“包含”关系。与相同疾病对应的多个药物之间为“相克”，“选择”或“并 列”关系，表示不能同时服用，选择部分服用，或需要共同服用。
步骤520，数据处理系统110(例如，信息查找模块1130)可以获取预设领域的原始文本和原始影像，从原始文本和原始影像中提取出多个实体数据，并确定多个实体数据之间的关联关系。
在一些实施例中，预设领域可以包括医疗领域，在一些实施例中，数据处理系统110获取原始文本和原始影像的方式可以包括：从书籍、报刊、网站获取医学常识的原始文本；从电子病历、化验单、诊断报告获取诊断相关的原始文本和原始影像；从医疗影像设备获取原始影像。本申请实施例对获取方式不做限定。
在一些实施例中，数据处理系统110可以将获取到的原始文本存储到MySQL数据表中，将获取到的原始影像以DICOM格式保存到影像数据库(例如，PACS，存储设备140，存储器220等)中，并将保存路径存储到MySQL数据表中。
在一些实施例中，对获取的原始文本可以先进行纠错处理，再进行存储。在一些实施例中，可以将获取的原始文本作为待处理文本，根据图12-图22的文本纠错方法对获取的原始文本进行纠错。
在一些实施例中，从书籍，报刊，网站，电子病历，化验单，诊断报告等中获取的原始文本和原始影像可以为非结构化数据，可以根据图23-图27的非结构化数据存储方法对原始文本和原始影像进行存储。
在一些实施例中，数据处理系统110获取到原始文本和原始影像后，可以从原始文本中提取出多个文本实体数据，并确定提取出的多个文本实体数据之间的关联关系；也可以从原始影像中提取出影像实体数据，并确定原始影像与影像实体数据之间的关联关系；还可以从原始文本和原始影像中提取出文本实体数据和影像实体数据，并确定文本实体数据和影像实体数据之间的关联关系。
步骤530，数据处理系统110(例如，信息查找模块1130)可以根据图谱结构、多个实体数据和多个实体数据之间的关联关系构建领域知识图谱。
在一些实施例中，在确定图谱结构、多个实体数据和多个实体数据之间的关联关系后，数据处理系统110可以根据多个实体数据之间的关联关系将多个实体数据组成多个三元组；然后，将多个三元组填充到图谱结构中，得到领域知识图谱，例如，将多个三元组存储到预先设置的图数据库中，得到领域知识图谱。
例如，以<实体，关系，实体>这样的三元组格式表示，可以将文本实体数据之间的关联关系表示为<XX肺炎(疾病)，使用，甲泼尼龙(药物)>，将文本实体数据与影像实 体数据之间的关联关系表示为<XX肺炎(疾病)，观察，6ff55da4-5454-4b81-a831-a25c7bf68f4d(特征影像)>，将影像实体数据支架的关联关系表示为<d3e616c8-1d39-4fd6-a2fe-e6a7e201732d(原始影像)，特征，6ff55da4-5454-4b81-a831-a25c7bf68f4d(特征影像)>。其中，d3e616c8-1d39-4fd6-a2fe-e6a7e201732d是某个原始影像的唯一标识，6ff55da4-5454-4b81-a831-a25c7bf68f4d为该原始影像的特征影像的唯一标识。之后，将上述三元组存储到Neo4j图数据库中，得到领域知识图谱，如图7所示。本申请实施例对图数据库不做限定。
如图7所示，原始影像d3e616c8-1d39-4fd6-a2fe-e6a7e201732d对应的特征影像为6ff55da4-5454-4b81-a831-a25c7bf68f4d。特征影像的影像描述为“两下肺纹理增多增粗”，诊断结果为“两下肺支气管炎”，对应的疾病为“支气管炎”。支气管炎的病因为“病毒感染”和/或“冷空气”，症状为“咳嗽”和/或“喘息”，治疗方案为“服用抗病毒、止咳化验药物”，用药为“达肺草”和/或“氨溴索口服液”，其中，“服用抗病毒、止咳化验药物”包含用药“达肺草”和/或“氨溴索口服液”。
上述实施例中，数据处理系统110确定领域知识图谱的图谱结构；获取预设领域的原始文本和原始影像，从原始文本和原始影像中提取出多个实体数据，并确定多个实体数据之间的关联关系；根据图谱结构、多个实体数据和多个实体数据之间的关联关系构建领域知识图谱。现有技术中，知识图谱通常只包含文本；而本申请实施例根据文本和影像建立了多模态的领域知识图谱。与现有技术相比，本申请实施例所建立的领域知识图谱更加全面、更加准确。
在一些实施例中，上述从原始文本和原始影像中提取出多个实体数据，并确定多个实体数据之间的关联关系的步骤，可以包括采用以下方式实现。
其中一种实体数据提取方式包括：对于原始文本，将原始文本输入到预先训练的分词模型中，得到分词模型输出的多个分词结果；将多个分词结果输入到预先训练的实体识别模型中，得到实体识别模型输出的特征文本，并将特征文本确定为实体数据。例如，原始文本中包括影像描述、诊断结果和疾病类型，可以将原始文本输入到预先训练的分词模型和实体识别模型中，得到原始文本中分别对应影像描述、诊断结果和疾病类型的特征文本。特征文本中可以包括表1中关于文本实体的属性的描述。
其中，分词模型和实体识别模型可以是终端自行训练的，也可以是由服务器训练好，终端再从服务器获取的。本申请实施例对此不做限定。
在一些实施例中，分词模型的训练过程可以包括：通过热词挖掘方式从海量文本中 提取出常用词及其标准表达；根据校验人员对常用词及其标准表达的校验结果，生成同义词表。接着，将同义词表转换成分词词典，再将分词词典和人工标注的分词数据作为输入训练分词模型。
在一些实施例中，可以将原始文本输入到分词模型中，得到分词模型输出的多个分词结果后，数据处理系统110针对分词结果进行N-Gram(N＝2,3,4…)的频次统计，并显示频次较高的K个N-Gram结果。接着，数据处理系统110接收校验人员针对频次较高的K个N-Gram结果输入的校验信息，根据校验信息确定是否需要将每个N-Gram结果合并成一个词汇，然后将合并得到的词汇更新到上述分词词典中。例如，一个原始文本“肾弓形动脉以规则的间距发出放射状的分支”经分词模型分词后得到的分词结果为“肾弓形动脉以规则的间距发出放射状的分支”，从中可以得到9个3-Gram{“肾弓形动脉”，“弓形动脉以”，“动脉以规则”，“规则的间距”，“的间距发出”，“间距发出放射”，“发出放射状”，“放射状的”，“状的分支”}，经人工确认“肾弓形动脉”是一个专业术语，因此将“肾弓形动脉”合并成“肾弓形动脉”加入到分词词典中。之后，可以根据更新后的分词词典重新训练分词模型。通过对分词结果进行N-Gram频次统计并人工校验，可以解决医疗领域较长的专业词汇被分割开的问题。
在一些实施例中，实体识别模型的训练过程可以包括：获取训练样本集，基于训练样本集进行识别模型的训练，得到上述实体识别模型。上述实体识别模型可以采用Word2Vec+BiLSTM+CRF结构。本申请实施例对实体识别模型的结构不做限定。
在一些实施例中，与上述实体数据提取方式对应的关联关系确定方式包括：将原始文本和多个特征文本输入到预先训练的关系提取模型中，得到关系提取模型输出的多个特征文本之间的关联关系。
在一些实施例中，关系提取模型可以是终端自行训练的；也可以是由服务器训练好，终端再从服务器获取的。本申请实施例对此不做限定。
在一些实施例中，数据处理系统110可以将原始文本和上述提取到的多个特征文本输入到关系提取模型中，关系提取模型根据原始文本的内容确定多个特征文本之间的关联关系，并输出该关联关系。上述关系提取模型可以采用门控循环单元(Gated Recurrent Unit，GRU)模型，本申请实施例对关系提取模型的结构不做限定。
可以理解地，数据处理系统110利用预先训练好的分词模型、实体识别模型和关系提取模型，可以快速、准确地从原始文本中提取出文本实体数据，并确定文本实体数据之间的关联关系，为构建领域知识图谱提供了大量文本依据。
另一种实体数据提取方式包括：对于原始影像，将原始影像输入到预先训练的识别模型中，得到识别模型输出的识别结果；根据识别结果从原始影像中截取出感兴趣区域的特征影像，并将特征影像确定为实体数据。
其中，识别结果用于指示原始影像中的感兴趣区域。例如，识别模型采用神经网络模型，将原始影像输入到神经网络模型后，神经网络模型输出原始影像中标有感兴趣区域识别框。接着，数据处理系统110将感兴趣区域识别框中的特征影像截取出来作为实体数据。本申请实施例对识别模型的结构不做限定。
在一些实施例中，与上述实体数据提取方式对应的关联关系确定方式包括：将特征影像存储到特征影像库中，并将特征影像和原始影像之间的关联关系存储到预先设置的数据表中。
在一些实施例中，从原始影像中截取出特征影像后，将特征影像存储到预先设置的特征影像库中，并建立“原始影像-特征影像”这样的关联关系。之后，将关联关系存储到MySQL数据表中。本申请实施例对数据表的结构不做限定。
可以理解地，数据处理系统110利用预先训练好的识别模型，可以快速、准确地从原始影像中截取出影像实体数据，并确定影像实体数据与原始影像之间的关联关系，为构建领域知识图谱提供了大量影像依据。
又一种实体数据提取方式：从原始文本中提取出原始影像对应的影像描述、诊断结果、疾病、诱因、症状、药物、治疗方案等文本中的至少一种，并确定为实体数据；从原始影像中提取出特征影像，并将特征影像确定为实体数据。
以影像描述和诊断结果为例，对于既包括原始文本和原始影像的原始数据，比如电子病历、诊断报告、影像报告等，数据处理系统110对原始影像进行识别，得到多个特征影像P＝{p 1,p 2,…,p n}，其中p n是一个特征影像。数据处理系统110对原始文本进行分解，得到了多个影像描述文本Q＝{q 1,q 2,…,q k}和多个诊断结果文本S＝{s 1,s 2,…,s m}。上述特征影像、影像描述文本和诊断结果均为实体数据。
在一些实施例中，与上述实体数据的提取方式对应的关联关系确定方式可以包括：建立特征影像、影像描述文本和诊断结果文本之间的关联关系。
在一些实施例中，数据处理系统110可以将影像描述文本和诊断结果文本存储到MySQL数据表中，将特征影像保存到特征影像库中，并将保存路径存储到MySQL数据表中。之后，建立关联关系“特征影像-[观察]-影像描述文本”，并用g(p i)＝q j表示，表明第i个特征影像与第j个影像描述文本对应。同时，还建立关联关系“特征影像-[结果]-诊断结果 文本”，并用g(p i)＝s j表示，表明第i个特征影像与第j个诊断结果文本对应。最后，将上述关联关系均存储到MySQL数据表中。
在一些实施例中，特征影像在保存到特征影像库中时，生成特征影像的唯一标识，建立特征影像与影像描述文本和诊断结果文本之间的关联关系时采用特征影像的唯一标识。
可以理解地，数据处理系统110从原始文本和原始影像中提取出实体数据，为构建领域知识图谱提供了大量文本依据和影像依据，使得构建出的领域知识图谱更加全面、更加准确。
在一些实施例中，数据处理系统110可以根据历史影像报告建立影像-文本对应关系。数据处理系统110可以根据目标影像和该影像-文本对应关系自动生成影像报告。在一些实施例中，数据处理系统110可以对历史影像报告进行识别(例如，使用图像识别模型)，得到历史影像报告中的影像的特征影像，并将特征影像存储到特征影像库中。数据处理系统110建立特征影像与历史影像报告的对应关系，存储在数据库中。在生成目标影像的影像报告时，数据处理系统110可以在特征影像库中查找与第一特征影像匹配的第二特征影像，然后通过特征影像与历史影像报告的对应关系，查找到与第二特征影像对应的历史影像报告。由于影像报告使用的模板一样，因此，数据处理系统110可以通过将目标影像和目标影像对应的患者基本信息放在历史影像报告的对应位置进行替换，生成目标影像的影像报告。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。例如，流程500可以由用户终端130执行，例如，CPU 340和/或GPU 330执行，和/或图11所示的一个或多个模块执行。
图8是根据本申请一些实施例所示的查找与目标影像对应的影像描述信息和诊断结果信息流程的示例性流程图。在一些实施例中，流程800可以由医疗系统100(如数据处理系统110)执行。例如，流程800可以由PACS，后处理工作站，医生工作站，或成像系统的服务器执行。又例如，流程800可以由处理器210执行。又例如，流程800可以由图11所示的一个或多个模块执行。在一些实施例中，流程800可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程800。下面呈现的流程800的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图8中示出的和下面描述的流程800的操作的顺序不旨在是限制性的。
如图8所示，上述基于预先构建的领域知识图谱和第二特征影像，查找与目标影像对应的影像描述信息和诊断结果信息的步骤，可以包括。
步骤810，数据处理系统110(例如，信息查找模块1130)可以在领域知识图谱中查找第二特征影像，并获取第二特征影像对应的影像描述集合和诊断结果集合。
在一些实施例中，影像描述集合可以由多个影像描述文本组成，诊断结果集合可以由多个诊断结果文本组成。
在一些实施例中，数据处理系统110从目标影像中截取出第一特征影像，根据第一特征影像在预先设置的特征影像库中进行查找，得到与第一特征影像匹配的第二特征影像，其中，第二特征影像可以是一个也可以是多个。之后，数据处理系统110在领域知识图谱中查找第二特征影像，与每个第二特征影像对应的影像描述文本和诊断结果文本可以是一个也可以是多个，这样，就获取到了多个影像描述文本和诊断结果文本。接着，将多个影像描述文本组成影像描述集合，将多个诊断结果文本组成诊断结果集合。
步骤820，数据处理系统110(例如，信息查找模块1130)可以对影像描述集合和诊断结果集合进行去重处理，得到去重后的影像描述文本和诊断结果文本，并根据去重后的影像描述文本和诊断结果文本确定影像描述信息和诊断结果信息。
在一些实施例中，由于影像描述集合中包括多个影像描述文本，这些影像描述文本可能存在重复描述的情况，因此需要对影像描述集合进行去重处理。同样地，对诊断结果结合也需要去重处理。去重处理的过程可以包括：对一个影像描述文本进行分词，得到分词结果；判断分词结果是否包含其他影像描述文本的文本内容，如果包含了另一个影像描述文本的内容，就将另一个影像描述文本去掉。诊断结果文本的去重处理可以参考上述过程，本申请实施例对此不做限定。
在一些实施例中，去重处理后，得到去重后的影像描述文本和诊断结果文本，将去重后的影像描述文本和诊断结果文本按照一定顺序排列，则得到了影像描述信息和诊断结果信息。
上述基于预先构建的领域知识图谱和第二特征影像，查找与目标影像对应的影像描述信息和诊断结果信息的过程中，终端在领域知识图谱中查找第二特征影像，并获取第二特征影像对应的影像描述集合和诊断结果集合；对影像描述集合和诊断结果集合进行去重处理，得到去重后的影像描述文本和诊断结果文本，并根据去重后的影像描述文本和诊断结果文本确定影像描述信息和诊断结果信息。本申请实施例中，终端利用领域知识图像进行影像描述信息和诊断结果信息的查找，由于领域知识图谱为包含文本和影像的多模态知识图谱， 因此，可以查找到更全面、更准确的影像描述信息和诊断结果信息。并且，终端对查找到的影像描述集合和诊断结果集合进行去重处理，可以避免影像报告中出现重复描述，使得影像报告更加简洁。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。例如，流程800可以由用户终端130执行，例如，CPU 340和/或GPU 330执行，和/或图11所示的一个或多个模块执行。
图9是根据本申请一些实施例所示的生成影像报告的示例性流程图。在一些实施例中，流程900可以由医疗系统100(如数据处理系统110)执行。例如，流程900可以由PACS，后处理工作站，医生工作站，或成像系统的服务器执行。又例如，流程900可以由处理器210执行。又例如，流程900可以由图11所示的一个或多个模块执行。在一些实施例中，流程900可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程900。下面呈现的流程900的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图9中示出的和下面描述的流程900的操作的顺序不旨在是限制性的。
如图9所示，在上述实施例的基础上，还可以包括：
步骤910，数据处理系统110(例如，信息查找模块1130)可以基于预先构建的领域知识图谱和第二特征影像，查找与第二特征影像对应的疾病信息、诱因信息、症状信息、药物信息、治疗方案信息。
在一些实施例中，领域知识图谱中与第二特征影像关联的还可以包括疾病信息，通过疾病信息可以通过关联查询推理得到疾病有关的诱因信息、症状信息、治疗方案信息和药物信息等，如图7所示。因此，在领域知识图谱中查找到第二特征影像所关联的影像描述文本和诊断结果后，还可以进一步查找到第二特征影像关联的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息等。可以理解地，与第二特征影像关联的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息可以作为目标影像对应的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息。
步骤920，数据处理系统110(例如，报告生成模块1140)可以根据目标影像、影像描述信息、诊断结果信息、疾病信息、诱因信息、症状信息、治疗方案信息和药物信息生成影像报告。
在一些实施例中，数据处理系统110可以预先设置影像报告模板，在获取到了目标影像对应的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息后，将这些信息填充到影像报告模板的对应位置中，自动生成影像报告。
例如，数据处理系统110可以获取影像报告模板，如图10所示，影像报告模板中可以包括患者基本信息模块1010，影像描述模块1020，以及诊断结果模块1030。数据处理系统110可以在获取到了目标影像对应的影像描述信息和诊断结果信息后，将这些信息填充到影像报告模板的对应的模块中，自动生成影像报告，如图10所示的影像报告1000。
上述实施例中，数据处理系统110基于预先构建的领域知识图谱和第二特征影像，查找与目标影像对应的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息；根据目标影像、影像描述信息、诊断结果信息、疾病信息、诱因信息、症状信息、治疗方案信息和药物信息生成影像报告。本申请实施例中，领域知识图谱不仅包含特征影像、影像描述文本和诊断结果文本，还包含疾病信息、诱因信息、症状信息、治疗方案信息和药物信息等，因此，与现有技术相比，本申请实施例结合知识图谱智能化的查询和推理功能丰富了影像报告的内容，可以为医生提供更多更全面的信息支持。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。例如，流程900可以由用户终端130执行，例如，CPU 340和/或GPU 330执行，和/或图11所示的一个或多个模块执行。
图11是根据本申请一些实施例所示的影像报告生成装置1100的示例性框图。在一些实施例中，数据处理系统110和/或用户终端130的硬件和/或软件可以基于影像报告生成装置1100实现。在一些实施例中，影像报告生成装置1100可以包括影像截取模块1110，影像查找模块1120，信息查找模块1130，以及报告生成模块1140。
影像截取模块1110，用于获取目标影像，并从目标影像中截取出第一特征影像；其中，第一特征影像为目标影像中感兴趣区域的影像。
影像查找模块1120，用于根据第一特征影像在预先设置的特征影像库中进行查找，得到与第一特征影像匹配的第二特征影像。
信息查找模块1130，用于基于预先构建的领域知识图谱和第二特征影像，查找出与第二特征影像对应的文本信息，例如，影像描述信息、诊断结果信息、疾病、诱因、症状、药物、治疗方案等中的至少一种。在一些实施例中，信息查找模块1130可以包括第一信息查找模块，用于基于预先构建的领域知识图谱和第二特征影像，查找出与第二特征影像对应 的影像描述信息和诊断结果信息。
报告生成模块1140，用于生成目标影像的影像报告。报告生成模块1140可以包括第一报告生成模块1140，用于根据目标影像、影像描述信息和诊断结果信息生成影像报告。
在其中一个实施例中，信息查找模块1130还包括：
结构确定模块，用于确定领域知识图谱的图谱结构；图谱结构包括实体类型和实体连接关系；
数据和关系提取模块，用于获取预设领域的原始文本和原始影像，从原始文本和原始影像中提取出多个实体数据，并确定多个实体数据之间的关联关系；
图谱构建模块，用于根据图谱结构、多个实体数据和多个实体数据之间的关联关系构建领域知识图谱。
在一些实施例中，实体类型包括原始影像、特征影像、影像描述、诊断结果、疾病、诱因、症状、药物、治疗方案中的至少一种。
在一些实施例中，实体连接关系包括观察、相克、并列、选择、使用、特征、结果、有关、并发症、包含、诱发、表现等中的至少一种。
在一些实施例中，上述数据和关系提取模块，具体用于对于原始文本，将原始文本输入到预先训练的分词模型中，得到分词模型输出的多个分词结果；将多个分词结果输入到预先训练的实体识别模型中，得到实体识别模型输出的特征文本，并将特征文本确定为实体数据。
在一些实施例中，上述数据和关系提取模块，具体用于将原始文本和多个特征文本输入到预先训练的关系提取模型中，得到关系提取模型输出的多个特征文本之间的关联关系。
在一些实施例中，上述数据和关系提取模块，具体用于对于原始影像，将原始影像输入到预先训练的识别模型中，得到识别模型输出的识别结果；识别结果用于指示原始影像中的感兴趣区域；从原始影像中截取出感兴趣区域的特征影像，并将特征影像确定为实体数据。
在一些实施例中，影像报告生成装置1100还包括：
影像存储模块2730，用于将特征影像存储到特征影像库中，并将特征影像和原始影像之间的关联关系存储到预先设置的数据表中。
在一些实施例中，上述数据和关系提取模块，具体用于从原始文本中提取出原始影 像对应的影像描述文本和诊断结果文本，并将影像描述文本和诊断结果文本确定为实体数据；从原始影像中提取出特征影像，并将特征影像确定为实体数据；建立特征影像、影像描述文本和诊断结果文本之间的关联关系。
在一些实施例中，上述图谱构建模块，具体用于根据多个实体数据之间的关联关系将多个实体数据组成多个三元组；将多个三元组存储到预先设置的图数据库中，得到领域知识图谱。
在一些实施例中，上述第一信息查找模块，具体用于在领域知识图谱中查找第二特征影像，并获取第二特征影像对应的影像描述集合和诊断结果集合；影像描述集合由多个影像描述文本组成，诊断结果集合由多个诊断结果文本组成；对影像描述集合和诊断结果集合进行去重处理，得到去重后的影像描述文本和诊断结果文本，并根据去重后的影像描述文本和诊断结果文本确定影像描述信息和诊断结果信息。
在一些实施例中，上述影像查找模块1120，具体用于将第一特征影像分别与特征影像库中的各特征影像进行相似度计算，得到多个相似度；将特征影像库中相似度满足预设条件的特征影像确定为第二特征影像。
在一些实施例中，信息查找模块1130还包括第二信息查找模块，用于基于预先构建的领域知识图谱和第二特征影像，查找与目标影像对应的疾病信息、诱因信息、症状信息、治疗方案信息和药物信息。
在一些实施例中，报告生成模块1140还包括第二报告生成模块，用于根据目标影像、影像描述信息、诊断结果信息、疾病信息、诱因信息、症状信息、治疗方案信息和药物信息生成影像报告。
关于影像报告生成装置1100的具体限定可以参见上文中对于影像报告生成方法(例如，流程400，流程500，流程800，流程900)的限定，在此不再赘述。上述影像报告生成装置1100中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。
在一些实施例中，本申请提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器(例如，数据处理系统110，用户终端130，处理器210，CPU  340，GPU 330，或图11的一个或多个模块)执行时实现流程400，流程500，流程800，流程900中的至少一个。
随着医疗信息系统在医疗行业的普及应用，医疗机构大部分都采用医疗信息系统进行医学数据的存储与编写，例如，检测报告、影像报告或诊断报告，而医学数据中医学文本数据占有很大的比例，如何对错误的医学文本数据进行校核是非常重要的环节。目前，对错误的医学文本数据进行校核的方式包括很多种，主要包括：采用具有检测错误文本功能的检测模型检测出错误的文本，然后人工修改错误的文本，得到纠正后的医学文本数据。然而，上述校核方法存在校核不准确的问题。针对上述技术问题，本申请提供一种能够有效提高纠错准确性和效率医学文本纠错的系统和方法。
图12A是根据本申请一些实施例所示的一种文本纠错方法的示例性流程图。在一些实施例中，流程1200可以由医疗系统100(如数据处理系统110)执行。例如，流程1200可以由图像存档与传输系统(PACS)，后处理工作站，手术系统，医院信息系统(HIS)，实验室信息系统(LIS)，放射科信息系统(RIS)，医生工作站，护士工作站，成像系统，住院管理系统，门急诊管理系统，或收费系统的服务器执行。又例如，流程1200可以由处理器210执行。又例如，流程1200可以由图22所示的一个或多个模块执行。在一些实施例中，流程1200可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程1200。下面呈现的流程1200的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图12A中示出的和下面描述的流程1200的操作的顺序不旨在是限制性的。
步骤1210，数据处理系统110可以对待处理文本进行错误检测，得到错误文本的类型和位置。
在一些实施例中，待处理文本为等待检错和纠错的文本。计算机设备可以在医院信息系统(Hospital Information System,HIS)、放射信息管理系统(Radiology information system,RIS)、电子病历系统、语音识别结果文本系统、问答系统等与医疗相关的系统中获取待处理文本，当然，计算机设备也可以从其它领域系统中获取待处理文本。错误文本的类型包括错字类型、缺失类型、无序类型和冗余类型。错误文本的位置表示错误的文本在待处理文本中所处的位置。文本可以是符号、数字、字母、单个文字、词、语句中的至少一种。
错误检测可以包括检测待处理文本中是否包括错误文本，和/或确定错误文本的类型和位置。在一些实施例中，计算机设备(例如，数据处理系统110)在得到待处理文本时， 可以将该待处理文本输入至已训练好的错误检测模型进行错误检测，得到该待处理文本中包含的错误文本的类型和错误文本在待处理文本中所处的位置。其中的错误检测模型可以采用神经网络模型或机器学习模型，例如，BiLSTM-CRF模型因其具有很好的检错功能得到广泛应用，本实施例对检测模型的选用不做限定。
示例性的，本实施例还提供了一种训练上述BiLSTM-CRF模型的方法，该方法包括：采用预设的序列标注方法对样本文本中的错误文本进行错误类型和位置的标注，得到标注文本，再将标注文本作为训练样本输入至初始的BiLSTM-CRF模型中进行训练，得到训练后的BiLSTM-CRF模型，使训练后的BiLSTM-CRF模型可以根据输入的待处理文本输出其中错误文本的位置和类型，以便上述使用。需要说明的是，上述序列标注方法可以采用任何类型的标注方法，例如，Bio序列标注方式或Bioes序列标注方式。
例如，图12B所示的示意图，其中样本文本为“这种材料表示吸烟引起了人们多么大的损害但是文章中的妻子是还有意识的，她还有活的意义”，将该待处理文本输入至Bio序列标注模型进行标注，得到标注后的样本文本为“这/O种/O材/O料/O表/B-S示/I-S吸/O烟/O引/O起/O了/O人/B-R们/I-R多/O么/O大/O的/O损/O害/O但/O是/O文/O章/O中/O的/O妻/O子/O是/B-W还/I-W有/O意/O识/O的/O，她/O还/O有/O活/B-W的/I-W意/O义/O”,其中，R表示冗余类型的文本，M表示缺失类型的文本，S表示错误类型的文本，W表示无序类型的文本，B表示错误文本的开始，I表示错误文本的内部，O表示错误文本的外部，关于采用什么符号进行标注可以由计算机设备预先设置，此处不做限定。计算机设备得到标注后的样本文本后，将该标注后的样本文本作为训练样本文本输入至待训练的BiLSTM-CRF模型进行训练。
步骤1220，数据处理系统110可以根据错误文本类型和纠错策略之间的对应关系，确定与错误文本的类型对应的目标纠错策略。
在一些实施例中，纠错策略表示对错误的文本进行纠错时采用的方法。错误文本类型和纠错策略之间的对应关系可以预先被存储在数据库中以便之后查看使用。
在一些实施例中，计算机设备基于前述步骤得到错误文本的类型时，可以进一步的根据错误文本的类型，根据错误文本类型和纠错策略之间的对应关系，确定与该错误文本的类型对应的目标纠错策略，以便之后根据该目标纠错策略指示的纠错方法对相应类型的错误文本进行纠错处理。例如，若错误文本的类型为错字类型，则对应的目标纠错策略包括拼音比较纠错策略和/或相似度比较纠错策略；若错误文本的类型为缺失类型或无序类型，则对应的目标纠错策略包括匹配度比较纠错策略；若错误文本的类型为冗余类型，则对应的目标 纠错策略包括删除纠错策略。
步骤1230，数据处理系统110可以根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理。
在一些实施例中，在计算机设备确定了与错误文本的类型对应的目标纠错策略后，即可按照该目标纠错策略指示的纠错方法，根据错误文本的位置在待处理文本中找到该错误文本，或者找到与该错误文本相关的文本，然后再针对该错误文本，或者与该错误文本相关的文本执行添加、或删除、或修正等不同的纠错操作，从而得到纠正后的文本。
本申请实施例所述的数据处理方法中，通过对待处理文本进行错误检测，得到错误文本的类型和位置，并根据错误文本类型和纠错策略之间的对应关系，确定与错误文本的类型对应的目标纠错策略，进一步的根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理。在上述纠错的数据处理方法中，由于每种类型的错误文本对应各自的目标纠错策略，实现了针对每种错误文本类型的纠错操作，相比于传统的无论任何类型的错误文本均采用大数据库获取用于纠错的文本导致的数据处理量大和纠错精度差的问题，本申请所述的纠错方法极大的提高了纠错的准确性，以及提高了数据处理速度。
在一些实施例中，当错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种时，提供了上述步骤1230的一种实现方式，如图13所述，上述步骤1230“根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理”，包括：
步骤1310，数据处理系统110可以根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本。
在一些实施例中，当错误文本的类型包括错字类型时，计算机设备可以通过比较错误文本与多个预设文本的拼音，确定与错误文本对应的至少一个候选文本，在一些实施例中，计算机设备也可以通过比较错误文本与多个预设文本之间的相似度，确定与错误文本对应的至少一个候选文本。当错误文本的类型包括缺失类型或无序类型时，计算机设备可以通过比较错误文本与多个预设文本的匹配度，确定与错误文本对应的至少一个候选文本。
在一些实施例中，可以为根据错误文本与预设文本的读音、拼写、文字、语义、上下文等中的至少一个确定错误文本与预设文本的相似度或匹配度。在一些实施例中，可以将相似度或匹配度大于阈值(例如，30％，40％，50％，60％，70％，80％，90％，95％等)的预设文本确定为候选文本。
步骤1320，数据处理系统110可以根据至少一个候选文本，对错误文本进行纠错处理。
当计算机设备基于上述步骤得到至少一个候选文本后，即可在该多个候选文本中任意选择一个候选文本对错误文本进行纠错，在一些实施例中，也可以按照预设规则，在该多个候选文本中选择一个候选文本对错误文本进行纠错，纠错的时候可以具体使用选出的候选文本替代错误文本，或者将选出的候选文本添加到待处理文本的相应位置上。上述预设规则可以由计算机设备预先根据应用需求确定。例如，预设规则可以包括选择相似度或匹配度最高的候选文本。
上述纠错的数据处理方法，实现了对错字类型、缺失类型、无序类型的错误文本进行纠错，且每种类型的错误文本对应不同的目标纠错策略，使在纠错过程中计算机设备可以根据具体类型的错误文本执行对应的纠错操作，极大的提高了纠错精度。
在一些实施例中，当上述错误文本的类型包括错字类型时，如图14所示，上述步骤1310“根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本”，包括：
步骤1410，根据错字类型的文本的位置在待处理文本中提取出错字类型的文本。
在一些实施例中，当计算机设备检测出的错误文本的类型为错字类型时，即可根据该错字类型的文本的位置在待处理文本中找到该错字类型的文本，并将该错字类型的文本提取出来，以便之后使用该错字类型的文本进行检索。
步骤1420，从预设的第一数据库中选取出与错字类型的文本同音，和/或，与错字类型的文本之间的相似度大于预设相似度阈值的文本作为候选文本。
在一些实施例中其中，第一数据库可以包括任一类型的数据库。预设相似度阈值可以由计算机设备预先根据纠错精度确定。具体地，当计算机设备得到错字类型的文本时，可以进一步的将该错字类型的文本作为目标文本在第一数据库中检索与该目标文本同音的文本，在一些实施例中，计算机设备也可以在第一数据库中检索与该目标文本谐音相同的文本，然后将检索到的文本作为候选文本；在一些实施例中，计算机设备也可以计算该目标文本与第一数据库中各文本之间的相似度，再将相似度大于预设相似度阈值的文本作为候选文本。在一些实施例中，计算机设备还可以计算该目标文本与第一数据库中各文本之间的编辑距离，再将编辑距离小于预设编辑距离阈值的文本作为候选文本，其中的预设编辑距离阈值可以由计算机设备预先根据纠错精度确定。
上述确定候选文本的方法，通过比较错误文本的拼音，或错误文本与第一数据库中的各文本的相似度即可确定出与错字类型的文本对应的候选文本，该方法简单实用，使之后基于候选文本进行纠错处理时可以提高纠错速度。
在一些实施例中，当上述错误文本的类型包括缺失类型时，如图15所示，上述步骤1310“根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本”，包括：
步骤1510，根据缺失类型的文本的位置在待处理文本中提取出与缺失类型的文本相邻的文本。
当计算机设备检测出的错误文本的类型为缺失类型时，即可根据该缺失类型的文本的位置在待处理文本中找到与该缺失类型的文本相邻的文本，并将该相邻的文本提取出来，以便之后使用该相邻的文本进行检索。
步骤1520，从预设的第二数据库中选取出与所述相邻的文本匹配的文本作为候选文本。
在一些实施例中其中，第二数据库可以包括医学领域专用的数据库，例如，医学知识图谱数据库。本实施例中的第二数据库与前述实施例中的第一数据库可以相同或不相同。
具体地，当计算机设备得到与缺失类型的文本相邻的文本时，可以进一步的将该相邻的文本作为目标文本在第二数据库中根据读音、拼写、文字、语义、上下文等中的至少一个检索与该目标文本匹配的文本，然后将检索到的文本作为候选文本。例如，待处理文本为“医院放科工作流程”，则检测出的缺失类型的文本为“放科”，与缺失类型的文本相邻的文本为“医院XX工作流程”。将“医院XX工作流程”作为目标文本在第二数据库中检索，得到“医院放射科工作流程”，则与该缺失类型的文本匹配的文本为“放射科”。上述确定候选文本的方法，通过比较错误文本与第一数据库中的各文本的匹配度即可确定出与缺失类型的错误文本对应的候选文本，该方法简单实用，使之后基于候选文本进行纠错处理时可以提高纠错速度。
在一些实施例中，当上述错误文本的类型包括无序类型时，如图16所示，上述步骤1310“根据目标纠错策略和错误文本的位置，确定与错误文本对应的至少一个候选文本”，包括：
步骤1610，根据无序类型的文本的位置在待处理文本中提取出无序类型的文本。
在一些实施例中，当计算机设备检测出的错误文本的类型为无序类型时，即可根据该无序类型的文本的位置在待处理文本中找到该无序类型的文本，并将该无序类型的文本提取出来，以便之后使用该无序类型的文本进行检索。
步骤1620，从预设的第三数据库中选取出与无序类型的文本包含相同字的文本作为候选文本。
在一些实施例中其中，第三数据库可以包括医学领域专用的数据库，例如，医学知识图谱数据库。本实施例中的第三数据库与前述实施例中的第二数据库可以相同，也可以不相同。
具体地，当计算机设备得到无序类型的文本时，可以进一步的将该无序类型的文本作为目标文本在第三数据库中检索与该目标文本包含相同字的文本，然后将检索到的文本作为候选文本。例如，待处理文本为“医院放科射工作流程”，则检测出的无序类型的文本为“放科射”，则与该无序类型的文本包含相同字的文本为“放射科”。上述确定候选文本的方法，通过在第三数据库中搜索与错误文本包含相同字的各文本即可确定出与无序类型的错误文本对应的候选文本，该方法简单实用，使之后基于候选文本进行纠错处理时可以提高纠错速度。
基于上述图12A至图16实施例所述的确定候选文本的方法，对应的，本申请还提供了上述1230的一种实现方式，如图17所示，所述方式包括：
步骤1710，将至少一个候选文本依次添加到待处理文本中对应位置上，对添加后的待处理文本的通顺程度进行打分。
当计算机设备基于前述图14至图16任一实施例所述的方法得到多个候选文本时，可以进一步的将这多个候选文本依次添加到待处理文本中对应位置上，得到多个添加后的待处理文本。然后计算机设备可以将多个添加后的待处理文本依次输入至已训练好的语言模型进行通顺程度的打分，分数高的表示添加后的待处理文本很通顺，分数低的表示添加后的待处理文本不通顺。
需要说明的是，上述语言模型用于分析添加后的待处理文本的通顺程度，并根据通顺程度的高低对添加后的待处理文本进行打分。在实际应用中，在训练上述语言模型时，可以在医学专业的数据库中获取训练样本数据，再将该训练样本数据输入至预设的预训练语言模型进行训练，得到训练后的语言模型。其中的预训练语言模型可以采用现有的BERT预训练语言模型，也可以采用其它类型的预训练语言模型，此处不做限定。
步骤1720，按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
当计算机设备对多个添加后的待处理文本进行打分后，可以进一步的按照分数从高到低的顺序对多个打分后的待处理文本进行排序，然后从中选择出分数最高的待处理文本作为纠正后的文本；在一些实施例中，计算机设备也可以先从多个打分后的待处理文本中选择出分数大于预设分数阈值的待处理文本，若该大于预设分数阈值的待处理文本的数量大于一 个时，则可以进一步的再从这选择出的多个待处理文本中选择出任意一个待处理文本，或者按照预设规则选择出一个待处理文本，最后将通过上述方法选择出的待处理文本作为纠正后的文本。
上述实施例所述的方法中，采用已训练好的语言模型对添加后的待处理文本进行通顺程度的打分，相比于仅是对错误的文本进行打分而造成分数不准确的方法，本实施例所述的方法由于结合了错误文本的上下文信息来进行打分，提高了根据分数选择出对应的候选文本进行纠错的准确性。
在一些个实施例中，当错误文本的类型包括冗余类型时，提供了上述1230的一种实现方式，如图18所述，上述步骤1230“根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理”，包括：
步骤1810，根据目标纠错策略和错误文本的位置，确定是否删除冗余类型的文本，若确定删除，则执行步骤1820，若确定不删除，则执行步骤1830。
具体地，当错误文本的类型包括冗余类型时，计算机设备可以将冗余类型的文本输入至已训练好的语言模型中进行检测，检测该冗余类型的文本是否通顺，在一些实施例中，计算机设备也可以在待处理文本中提取包含该冗余类型的文本的文本，然后将提取出的文本输入至已训练好的语言模型中进行检测，检测该提取出的文本是否通顺，若通顺则说明该冗余类型的文本实际上在待处理文本中不是冗余的文本，若不通顺则说明该冗余类型的文本实际上在待处理文本中是冗余的文本。在一些实施例中，训练好的语言模型可以输出“通顺”或“不通顺”的结果，例如，训练好的语言模型可以输出“1”表示“通顺”，输出“0”表示“不通顺”。在一些实施例中，训练好的语言模型可以输出表示通顺程度的数值，例如，训练好的语言模型可以输出“90％”表示通顺程度为90％。通顺程度大于或等于一定阈值(例如，30％，40％，50％，60％，70％，80％，90％，95％等)的文本可以认为通顺，通顺程度小于该阈值的文本可以认为不通顺。
需要说明的是，上述语言模型用于分析冗余类型的文本或包含冗余类型的文本的文本是否通顺，并输出包含通顺或不通顺的分析结果。在实际应用中，在训练上述语言模型时，可以在医学专业的数据库中获取训练样本数据，再将该训练样本数据输入至预设的预训练语言模型进行训练，得到训练后的语言模型。其中的预训练语言模型可以采用现有的BERT预训练语言模型，也可以采用其它类型的预训练语言模型，此处不做限定。
步骤1820，删除待处理文本中的错误文本。
本实施例涉及的是计算机设备确定删除冗余类型的文本的应用场景，在此场景中， 计算机设备直接执行删除待处理文本中的冗余文本的操作。
步骤1830，不执行删除操作。
本实施例涉及的是计算机设备确定不删除冗余类型的文本的应用场景，在此场景中，计算机设备不执行删除操作。
在一些实施例中进一步地，如图19所示，上述1810“根据目标纠错策略和错误文本的位置，确定是否删除冗余类型的文本”，包括：
步骤1910，根据冗余类型的文本的位置在待处理文本中提取出包含冗余类型的文本的文本。
在一些实施例中，当计算机设备检测出的错误文本的类型为冗余类型时，即可根据该冗余类型的文本的位置在待处理文本中找到该冗余类型的文本，并将包含该冗余类型的文本的文本提取出来，以便之后使用该包含冗余类型的文本的文本进行检索。例如，若待处理文本为“医疗放放射科工作流程”，其中冗余类型的文本为“放放”，然后提取出的包含该冗余类型的文本的文本可以是“放放射科”，也可以是“医疗放放射科”，至于包含冗余类型的文本的文本的长度可以由计算机设备预先确定，此处不做限定。
步骤1920，检测包含冗余类型的文本的文本是否通顺，若不通顺，则执行步骤1930，若通顺，则执行步骤1940。
步骤1930，确定删除冗余类型的文本。
步骤1940，确定不删除冗余类型的文本。
在一些实施例中，当计算机设备得到包含冗余类型的文本的文本时，可以将该文本输入至已训练好的语言模型中进行检测，检测该提取出的文本是否通顺，若不通顺，则确定删除冗余类型的文本，若通顺，则确定不删除冗余类型的文本。
上述实施例所述的方法中，采用已训练好的语言模型对包含冗余类型的文本的文本进行分析，判断该文本是否通顺，相比于仅是对冗余类型的文本进行分析而造成判断不准确的方法，本实施例所述的方法由于结合了冗余类型的文本的上下文信息来进行判断是否需要被删除，提高了对冗余类型的文本进行纠错的准确性。
综合上述所有实施例，本申请还提供了一种数据处理方法，如图20所示，该数据处理方法包括：
步骤2001，对待处理文本进行错误检测，得到错误文本的类型和位置。
步骤2003，判断错误文本的类型。当错误文本的类型包括错字类型时，执行步骤2004-2005、2010-2011；当错误文本的类型包括缺失类型时，执行步骤2006-2007、2010- 2011；当错误文本的类型包括无序类型时，执行步骤2008-2009、2010-2011；当错误文本的类型包括冗余类型时，执行步骤2012-2015。
步骤2004，根据错字类型的文本的位置在待处理文本中提取出错字类型的文本。
步骤2005，从预设的第一数据库中选取出与错字类型的文本同音，和/或，与错字类型的文本之间的相似度大于预设相似度阈值的文本作为候选文本。
步骤2006，根据缺失类型的文本的位置在待处理文本中提取出与缺失类型的文本相邻的文本。
步骤2007，从预设的第二数据库中选取出与所述相邻的文本匹配的文本作为候选文本。
步骤2008，根据无序类型的文本的位置在待处理文本中提取出无序类型的文本。
步骤2009，从预设的第三数据库中选取出与无序类型的文本包含相同字的文本作为候选文本。
步骤2010，将至少一个候选文本依次添加到待处理文本中对应位置上，对添加后的待处理文本的通顺程度进行打分。
步骤2011，按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
步骤2012，根据冗余类型的文本的位置在待处理文本中提取出包含冗余类型的文本的文本。
步骤2013，检测包含冗余类型的文本的文本是否通顺，若不通顺，则执行步骤2014，若通顺，则执行步骤2015。
步骤2014，确定删除冗余类型的文本，删除待处理文本中的错误文本。
步骤2015，确定不删除冗余类型的文本，不执行删除操作。
在另一个实施例中，检测出待处理文本的同一位置包含多个类型的错误，可以按照设定好的顺序依次进行以上处理策略。
上述实施例所述方法的每一个步骤在前述实施例中已进行说明，详细内容请参见前述说明，此处不赘述。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。例如，上述流程可以由用户终端130执行，例如，CPU 340和/或GPU 330执行，和/或图22所示的一个或多个模块执行。
综合上述所有实施例，本申请还提供了一种纠错系统2100，如图21所示，该纠错系统包括：错误检测模型2110、筛选召回模型2120、纠错处理模型2130。错误检测模型2110的输出端与筛选召回模型2120的输入端连接，筛选召回模型2120的输出端与纠错处理模型2130的输入端连接。在一些实施例中其中，错误检测模型2110用于对输入的文本进行错误文本的检测，输出错误文本的类型和位置；当错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种时，筛选召回模型2120用于根据错误检测模型输出的错误文本的类型选择与该类型对应的纠错处理策略，然后根据选择出的纠错处理策略和错误文本的位置，在预设的数据库中确定与错误文本对应的多个候选文本，然后纠错处理模型对多个候选文本进行通顺程度的打分，并根据分数最高的候选文本对待处理文本进行纠错；当错误文本的类型包括冗余类型时，筛选召回模型2120用于检测冗余类型的文本，或者包括冗余类型的文本的文本是否通顺，以确定是否需要删除该冗余类型的文本，然后纠错处理模型2130在筛选召回模型2120判定冗余类型的文本(包括冗余类型的文本的文本)是通顺的情况下，不执行删除操作；在筛选召回模型2120判定冗余类型的文本(包括冗余类型的文本的文本)不通顺的情况下，在待处理文本中删除冗余类型的文本。前述数据处理方法应用于上述纠错系统，用于对文本进行纠错处理，关于该纠错系统中包含的各模型或各部件的功能和实现方法请参见前述实施例说明，此处不赘述说明。在一些实施例中，数据处理系统110和/或用户终端130的硬件和/或软件可以基于纠错系统2100实现。
在一些实施例中，错误文本的类型还可以包括内容错误。例如，影像报告中可以包括目标影像和文本。如果一段文本的内容与其它文本或者目标影像不对应，则该文本为错误文本，错误类型为内容错误。例如，如图10所示，影像报告中的检查项目为“头颅CT轴位平扫”，影像报告中的影像描述和诊断意见为关于胸腔的描述，则，影像报告中的影像描述和诊断意见文本为错误文本，错误类型为内容错误。又例如，影像报告中的目标影像为头部影像，影像报告中的影像描述和诊断意见为关于胸腔的描述，则，影像报告中的影像描述和诊断意见文本为错误文本，错误类型为内容错误。
作为示例，可以将影像报告输入到预先训练的分词模型中，得到分词模型输出的影像报告中的文本的多个分词结果。将多个分词结果输入到预先训练的文字识别模型中，确定分词结果中的不相关的分词。可以将文字识别模型识别的不相关的分词所在地语句确定为错误文本，并确定错误文本的类型和位置。作为又一示例，可以提取影像报告中的目标影像，将提取的目标影像输入图像识别模型中，得到模型输出的影像特征。将影像特征和上述多个分词结果输入到预先训练的文字识别模型中，确定分词结果中与影像特征不相关的分词。可 以将文字识别模型识别的不相关的分词所在地语句确定为错误文本，并确定错误文本的类型和位置。在一些实施例中，识别出内容错误文本后，可以利用图4-图11中描述的方法重新在知识谱图中查找用于替换错误文本的文本描述，更新影像报告。
在一个实施例中，如图22所示，提供了一种数据处理装置2200，包括：
检测模块2210，用于对待处理文本进行错误检测，得到错误文本的类型和位置；
确定模块2220，用于根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
纠错模块2230，用于根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
关于数据处理装置2200的具体限定可以参见上文中对于数据处理方法的限定，在此不再赘述。上述数据处理装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。在一些实施例中，数据处理系统110和/或用户终端130的硬件和/或软件可以基于数据处理装置2200实现。
在一些实施例中，本申请提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器(例如，数据处理系统110，用户终端130，处理器210，CPU 340，GPU 330，图21的纠错系统2100，或图22的一个或多个模块)执行时实现流程1200-流程2000中的至少一个。
随着医疗信息化的不断发展和创新，产生了大量的医疗数据，而这些医疗数据中大部分数据为非结构化数据，而且随着医疗设备技术的发展和检查手段的进步，在提升医疗水平的同时也会导致更大量的非结构化数据的产生，因此，对这些非结构化的医疗数据的存储就显得尤为重要。传统技术中，对于非结构化的医疗数据的存储主要是根据非结构化的医疗数据的数据标识，采用定制接口以及备份软件加磁带库或者虚拟磁带库的方法进行存储，但是，医疗数据通常是海量数据，在使用该方法进行非结构化医疗数据的存储时仅扫描一遍文件系统就需要等待漫长的时间。因此，传统的非结构化医疗数据的存储方法，存在存储效率较低的问题。本申请针对上述技术问题，提供一种能够提高非结构化医疗数据的存储效率的非结构化数据存储方法、装置、计算机设备和存储介质。
图23是根据本申请一些实施例所示的一种非结构化数据存储方法的示例性流程图。在一些实施例中，流程2300可以由医疗系统100(如数据处理系统110)执行。例如，流程 2300可以由图像存档与传输系统(PACS)，后处理工作站，手术系统，医院信息系统(HIS)，实验室信息系统(LIS)，放射科信息系统(RIS)，医生工作站，护士工作站，成像系统，住院管理系统，门急诊管理系统，或收费系统的服务器执行。又例如，流程2300可以由处理器210执行。又例如，流程2300可以由图27所示的一个或多个模块执行。在一些实施例中，流程2300可以以程序或指令的形式存储在存储装置(如存储设备140)中，当医疗系统100执行该程序或指令时，可以实现流程2300。下面呈现的流程2300的操作示意图是说明性的。在一些实施例中，可以利用一个或以上未描述的附加操作和/或未讨论的一个或以上操作来完成该过程。另外，图23中示出的和下面描述的流程2300的操作的顺序不旨在是限制性的。
步骤2310，获取待存储的非结构化医疗数据。
在一些实施例中，非结构化数据是指不能用二维逻辑在数据库表中进行表现的数据，本申请中，非结构化医疗数据可以指与医疗系统100有关的办公文档、文本、图片、XML、HTML、各类报表、音频、视频等，例如，计算机断层成像(Computed Tomography，CT)图像、磁共振(Magnetic Resonance Imaging，MRI)图像、PET图像、分子影像(Molecular Imaging，MI)、X射线(X-ray，XR)图像、超声图像等非结构化的医学影像数据，影像报告，诊断报告，化验单，申请单，医疗系统100从外部设备获取的文本、图片、音频、视频等。
具体地，计算机设备(例如，数据处理系统110)获取待存储的非结构化医疗数据。在一些实施例中，计算机设备可以从医疗系统100的组件中获取待存储的非结构化医疗数据，例如，从PACS(Picture Archiving and Communication Systems，影像归档和通信系统)服务器中获取待存储的非结构化医疗数据，也可以从医学影像设备中实时地获取待存储的非结构化医疗数据。在一些实施例中，计算机设备可以从医疗系统100的外部设备中获取待存储的非结构化医疗数据。在一些实施例中，计算机设备获取到待存储的非结构化医疗数据后可以首先对获取的待存储的非结构化医疗数据进行预处理，去除掉获取的待存储的非结构化医疗数据中的冗余数据等。
步骤2320，对非结构化医疗数据进行解析，得到非结构化医疗数据对应的结构化医疗数据，以及非结构化医疗数据与结构化医疗数据之间的对应关系(索引信息)。
结构化数据可以指由二维表结构来逻辑表达和实现的数据。可以通过离线解析或在线解析对非结构化数据进行解析。具体地，计算机设备对上述获取的待存储的非结构化医疗数据进行解析，得到该非结构化医疗数据对应的结构化医疗数据，以及该非结构化医疗数据 与得到的结构化医疗数据之间的对应关系。在一些实施例中，计算机设备可以采用预设的影像解析方法对待存储的非结构化医疗数据进行解析，得到待存储的非结构化医疗数据对应的结构化医疗数据，也可以采用预设的文本提取方法对待存储的非结构化医疗数据进行解析，得到待存储的非结构化医疗数据对应的结构化医疗数据，进而根据上述待存储的非结构化医疗数据和得到的结构化医疗数据，得到非结构化医疗数据与结构化医疗数据之间的对应关系。在一些实施例中，计算机设备可以根据对待存储的非结构化医疗数据进行解析时的解析路径，得到待存储的非结构化医疗数据和结构化医疗数据之间的对应关系。需要说明的是，不同类型的非结构化数据对应的解析方法也是不同的，相应地，不同类型的非结构化数据所对应的非结构化医疗数据与结构化医疗数据之间的对应关系也是不同的，在一些实施例中，计算机设备可以根据上述待存储的非结构化医疗数据的类型确定该待存储的非结构化医疗对应的解析方法，从而对该非结构化医疗数据进行解析，得到该非结构化医疗数据对应的结构化医疗数据。在一些实施例中，计算机设备还可以将不同类型的非结构化数据所对应的非结构化医疗数据与结构化医疗数据之间的对应关系按需进行最大集合标准处理，也就是说，计算机设备可以根据用户的需求，将非结构化医疗数据与结构化医疗数据之间的对应关系进行最大集合标准处理，将非结构化医疗数据与结构化医疗数据之间的对应关系存储到不同的集合中。
步骤2330，将结构化医疗数据和对应关系存储到预设的第四数据库中，将非结构化医疗数据存储到预设的第五数据库中。
具体地，计算机设备将上述得到的结构化医疗数据和非结构化医疗数据与结构化医疗数据之间的对应关系存储到预设的第四数据库中，将待存储的非结构化数据存储到预设的第五数据库中。在一些实施例中，预设的第四数据库可以为DB数据库，预设的第五数据库可以为对象存储数据库。在一些实施例中，计算机设备可以实时地将结构化医疗数据和得到的非结构化医疗数据与结构化医疗数据之间的对应关系存储到预设的第四数据库中，也可以按照预设的时间间隔例如5秒等，将结构化医疗数据和得到的非结构化医疗数据与结构化医疗数据之间的对应关系存储到预设的第四数据库中，相应地，计算机设备可以实时地将待存储的非结构化医疗数据存储到预设的第五数据库中，也可以按照上述预设的时间间隔将待存储的非结构化医疗数据存储到预设的第五数据库中。
上述非结构化数据存储方法中，由于对非结构化医疗数据进行解析的过程十分简单，从而计算机设备能够快速准确地得到待存储的非结构化医疗数据对应的结构化医疗数据，以及该非结构化医疗数据与得到的结构化医疗数据之间的对应关系，这样计算机设备能 够快速地将得到的结构化医疗数据和得到的对应关系存储到预设的第四数据库中，将待存储的非结构化医疗数据存储到预设的第五数据库中，从而提高了对待存储的非结构化医疗数据的存储效率。
在上述对待存储的非结构化医疗数据进行解析，得到待存储的非结构化医疗数据对应的结构化医疗数据以及非结构化医疗与结构化医疗数据之间的对应关系的场景中，在一个实施例中，如图24所示，上述步骤2320，包括：
步骤2410，采用预设的解析规则，对非结构化医疗数据进行解析，得到非结构化医疗数据对应的结构化医疗数据。
具体地，计算机设备采用预设的解析规则，对待存储的非结构化医疗数据进行解析，得到待存储的非结构化医疗数据对应的结构化医疗数据。在一些实施例中，预设的解析规则可以是预设的文字解析规则，也可以是预设的影像解析规则，也可以是预设的报告解析规则。示例性地，计算机设备可以采用预设的文字解析规则，对待存储的非结构化数据中的文字进行解析，得到该待存储的非结构化数据对应的结构化医疗数据。
步骤2420，根据非结构化医疗数据的标识和结构化医疗数据的标识，建立非结构化医疗数据和结构化医疗数据之间的对应关系。
具体地，计算机设备根据上述待存储的非结构化医疗数据的标识和得到的结构化医疗数据的标识，建立上述待存储的非结构化医疗数据和得到的结构化医疗数据之间的对应关系。示例性地，以待存储的非结构化医疗数据的标识为A，得到的非结构化医疗数据对应的结构化医疗数据的标识为B为例，则建立的待存储的非结构化医疗数据和得到的结构化医疗数据之间的对应关系为：A-B。
本实施例中，计算机设备采用预设的解析规则，对待存储的非结构化医疗数据进行解析的过程十分地简单，从而使计算机设备能够快速地得到待存储的非结构化医疗数据对应的结构化医疗数据，提高了计算机设备得到待存储的非结构化医疗数据对应的结构化医疗数据的效率，进而可以使计算机设备根据待存储的非结构化医疗数据的标识和得到的结构化医疗数据的标识，快速地建立待存储的非结构化医疗数据和得到的结构化医疗数据之间的对应关系，从而提高了计算机设备得到待存储的非结构化医疗数据和得到的结构化医疗数据之间的对应关系的效率。
在上述采用预设的解析规则，对待存储的非结构化医疗数据进行解析的场景中，在一个实施例中，上述预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
具体地，上述预设的解析规则包括医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则，也就是，计算机设备可以采用医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则，对待存储的非结构化医疗数据进行解析，得到该待存储的非结构化医疗数据对应的结构化医疗数据。在一些实施例中，计算机设备可以采用医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则对待存储的非结构化医疗数据进行实时地解析，也可以采用医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则对待存储的非结构化医疗数据在后台解析后，将得到的对应的结构化医疗数据传给计算机设备。示例性地，医学影像文字识别规则通过预先配置各个文字代表的含义，从而识别出待存储的非结构化医疗数据中的各文字，解析出各文字代表的含义，得到待存储的非结构化医疗数据对应的结构化医疗数据，例如，医学影像文字识别规则可以对申请单进行解析，得到申请单中的各文字代表的含义，从而得到该申请单对应的结构化医疗数据；DICOM文件解析规则即医学数字成像和通信规则，是医学图像和相关信息的国际标准(ISO 12052)，它定义了质量能满足临床需要的可用于数据交换的医学图像格式，通过该国际标准可以对待存储的非结构化医疗数据进行解析，得到对应的结构化医疗数据；XML报告解析规则通过配置自定义的解析规则，对待存储的非结构化医疗数据进行解析，例如，对CDA临床文档架构进行解析，在一些实施例中，CDA提供了一个能够表达所有可能文档的通用架构，所有的CDA文档都用XML编码表达，因此，可以通过XML报告解析规则对CDA文档进行解析，得到对应的结构化医疗数据。
本实施例中，预设的解析规则包括医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则，而医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则涵盖的范围较广，从而能够使计算机设备对待存储的非结构化医疗数据进行全面地解析，扩大了预设的解析规则的使用场景。
在上述获取待存储的非结构化医疗数据的场景中，计算机设备可以通过被动接收的方式获取待存储的非结构化医疗数据，也可以通过主动获取的方式获取待存储的非结构化医疗数据，下面详细介绍采用这两种方法的具体实现方式：
第一种：若计算机设备是通过被动接收的方式获取的待存储的非结构化医疗数据，则上述步骤2310，包括：通过预设的应用程序接口，接收对应的医疗设备发送的待存储的非结构化医疗数据。
具体地，计算机设备通过预设的应用程序接口，接收对应的医疗设备发送的待存储的非结构化医疗数据。示例性地，以待存储的非结构化医疗数据为CT数据为例，则计算机 设备可以通过与CT设备间的应用程序接口，接收CT设备发送的待存储的非结构化医疗数据。
本实施例中，计算机设备通过预设的应用程序接口，能够完整地接收到对应的医疗设备发送的待存储的非结构化医疗数据，从而保证了计算机设备接收到的待存储的非结构化医疗数据的完整性。
第二种：若计算机设备是通过主动获取的方式获取的待存储的非结构化医疗数据，则上述步骤2310，包括：根据目标地址标识，在目标地址标识对应的医疗设备中进行查找，得到待存储的非结构化医疗数据。
具体地，计算机设备根据目标地址标识，在目标地址标识对应的医疗设备中进行查找，得到待存储的非结构化医疗数据。示例性地，以目标地址标识为MR设备的目标地址标识为例，则计算机设备获取待存储的非结构化医疗数据时可以根据该MR设备的目标地址标识，在该MR设备的目标地址标识对应的MR设备中进行查找，得到待存储的非结构化医疗数据。
本实施例中，计算机设备根据目标地址标识，能够在目标地址标识对应的医疗设备中快速地进行查找，从而能够快速地得到待存储的非结构化医疗数据，提高了得到待存储的非结构化医疗数据的效率。
在一些场景中，将待存储的非结构化医疗数据和待存储的非结构化医疗数据对应的结构化医疗数据进行存储后，还需要对存储的非结构化医疗数据和存储的结构化医疗数据进行查询。在一个实施例中，如图25所示，上述方法还包括：
步骤2510，接收查询请求；查询请求携带有待查询数据的数据标识。
具体地，计算机设备接收查询请求，其中，该查询请求中携带有待查询数据的数据标识。在一些实施例中，待查询数据的数据标识可以是非结构化医疗数据的数据标识，也可以是结构化医疗数据的数据标识。在一些实施例中，计算机设备接收到的查询请求可以是一个查询请求，也可以是多个查询请求。
步骤2520，基于查询请求，根据数据标识和对应关系，分别在第四数据库和第五数据库进行查询，得到数据标识对应的结构化医疗数据和非结构化医疗数据。
具体地，计算机设备基于接收到的查询请求，根据查询请求中携带的待查询数据的数据标识和存储的非结构化医疗数据与存储的结构化医疗数据之间的对应关系，分别在上述第四数据库和上述第五数据库中进行查询，得到查询请求中携带的待查询数据的数据标识对应的结构化医疗数据和非结构化医疗数据。在一些实施例中，计算机设备可以基于接收到的 查询请求，根据查询请求中携带的待查询数据的数据标识在上述第四数据库中进行查询，得到查询请求中携带的待查询数据的数据标识对应的结构化医疗数据，根据查询到的上述数据标识对应的结构化医疗数据和上述待存储的非结构化医疗数据与对应的结构化医疗数据之间的对应关系，在上述第五数据库中进行查询，得到查询请求中携带的待查询数据的数据标识对应的非结构化医疗数据。
本实施例中，计算机设备基于接收到的查询请求，能够根据该查询请求中携带的待查询数据的数据标识和待存储的非结构化医疗数据与对应的结构化医疗数据之间的对应关系，分别在第四数据库和第五数据进行准确地查询，准确地得到查询请求中携带的待查询数据的数据标识对应的结构化医疗数据和非结构化医疗数据，提高了得到的查询请求对应的结构化医疗数据和非结构化医疗数据的准确度；另外，根据查询请求中携带的待查询数据的数据标识和待存储的非结构化医疗数据与对应的结构化医疗数据之间的对应关系，能够快速地在第四数据库和第五数据库中查询到查询请求中携带的待查询数据的数据标识对应的结构化医疗数据和非结构化医疗数据，提高了得到查询请求中携带的待查询数据的数据标识对应的结构化医疗数据和非结构化医疗数据的效率。
为了便于本领域技术人员的理解，以下对本申请提供的非结构化数据存储方法进行详细介绍，如图26A和图26B所示，该方法可以包括：
S1，通过预设的应用程序接口，接收对应的医疗设备发送的待存储的非结构化医疗数据，或者，根据目标地址标识，在目标地址标识对应的医疗设备中进行查找，得到待存储的非结构化医疗数据。
S2，采用预设的解析规则，对非结构化医疗数据进行解析，得到非结构化医疗数据对应的结构化医疗数据；预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
S3，根据非结构化医疗数据的标识和结构化医疗数据的标识，建立非结构化医疗数据和结构化医疗数据之间的对应关系。
S4，将结构化医疗数据和对应关系存储到预设的第四数据库中，将非结构化医疗数据存储到预设的第五数据库中。
S5，接收查询请求；查询请求携带有待查询数据的数据标识。
S6，基于查询请求，根据数据标识在第四数据库中进行查询，得到数据标识对应的结构化医疗数据。
S7，根据数据标识对应的结构化医疗数据和对应关系，在第五数据库中进行查询， 得到数据标识对应的非结构化医疗数据。
需要说明的是，针对上述S1-S7中的描述可以参见上述实施例中相关的描述，且其效果类似，本实施例在此不再赘述。
应当注意的是，上述描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。
在一个实施例中，如图27所示，提供了一种非结构化数据存储装置2700，包括：获取模块2710、解析模块2720和存储模块2730，其中：
获取模块2710，用于获取待存储的非结构化医疗数据。
解析模块2720，用于对非结构化医疗数据进行解析，得到非结构化医疗数据对应的结构化医疗数据，以及非结构化医疗数据与结构化医疗数据之间的对应关系。
存储模块2730，用于将结构化医疗数据和对应关系存储到预设的第四数据库中，将非结构化医疗数据存储到预设的第五数据库中。
本实施例提供的非结构化数据存储装置2700，可以执行上述方法实施例，其实现原理和技术效果类似，在此不再赘述。
在上述实施例的基础上，在一些实施例中，上述解析模块2720包括：解析单元和建立单元，其中：
解析单元，用于采用预设的解析规则，对非结构化医疗数据进行解析，得到非结构化医疗数据对应的结构化医疗数据。
建立单元，用于根据非结构化医疗数据的标识和结构化医疗数据的标识，建立非结构化医疗数据和结构化医疗数据之间的对应关系。
在一些实施例中，预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
本实施例提供的非结构化数据存储装置，可以执行上述方法实施例，其实现原理和技术效果类似，在此不再赘述。
在上述实施例的基础上，在一些实施例中，上述获取模块2710包括：第一获取单元，其中：
第一获取单元，用于通过预设的应用程序接口，接收对应的医疗设备发送的待存储的非结构化医疗数据。
本实施例提供的非结构化数据存储装置，可以执行上述方法实施例，其实现原理和 技术效果类似，在此不再赘述。
在上述实施例的基础上，在一些实施例中，上述获取模块2710包括：第二获取单元，其中：
第二获取单元，用于根据目标地址标识，在目标地址标识对应的医疗设备中进行查找，得到待存储的非结构化医疗数据。
本实施例提供的非结构化数据存储装置，可以执行上述方法实施例，其实现原理和技术效果类似，在此不再赘述。
在上述实施例的基础上，在一些实施例中，上述装置还包括：接收模块和查询模块，其中：
接收模块，用于接收查询请求；查询请求携带有待查询数据的数据标识。
查询模块，用于基于查询请求，根据数据标识和对应关系，分别在第四数据库和第五数据库进行查询，得到数据标识对应的结构化医疗数据和非结构化医疗数据。
本实施例提供的非结构化数据存储装置，可以执行上述方法实施例，其实现原理和技术效果类似，在此不再赘述。
在上述实施例的基础上，在一些实施例中，上述查询模块包括：第一查询单元和第二查询单元，其中：
第一查询单元，用于基于查询请求，根据数据标识在第四数据库中进行查询，得到数据标识对应的结构化医疗数据。
第二查询单元，用于根据数据标识对应的结构化医疗数据和对应关系，在第五数据库中进行查询，得到数据标识对应的非结构化医疗数据。
本实施例提供的非结构化数据存储装置，可以执行上述方法实施例，其实现原理和技术效果类似，在此不再赘述。
关于非结构化数据存储装置的具体限定可以参见上文中对于非结构化数据存储方法的限定，在此不再赘述。上述非结构化数据存储装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一些实施例中，本申请提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器(例如，数据处理系统110，用户终端130，处理器210，CPU 340，GPU 330，或图27的一个或多个模块)执行时实现流程2300，2400，2500，2600-2， 2600-2中的至少一个。
图28是根据本申请一些实施例所示的医疗系统工作流2800的示例性流程图。流程2800包括医疗系统100中进行影像检查的工作流。
如图28所示，流程2800可以包括下述步骤。
步骤2810，获取申请单。
申请单可以为记载有拍摄医学影像请求的文件。在一些实施例中，申请单的格式可以为文档、图片等。在一些实施例中，拍摄医学影像请求可以以文字的形式记载在申请单上。在一些实施例中，拍摄医学影像请求可以包括与拍摄医学影像相关的信息，例如，病人相关信息、拍摄相关信息等。示例性病人相关信息可以包括姓名、年龄、性别等。示例性拍摄相关信息可以包括感兴趣区域(例如，脑部，胸部，肺部，腹部，心脏，骨骼等)、影像类型(例如，CT影像、MR影像、PET影像、X光片、超声影像等)、拍摄时间等。
在一些实施例中，申请单可以存储在，例如，医院信息系统(HIS)，或者与影像拍摄相关的系统中，例如，放射科信息系统(RIS)，成像系统。申请单为非结构化数据，可以根据图23-图28的非结构化数据存储方法对申请单进行存储。例如，对所述申请单进行解析，得到所述申请单对应的结构化医疗数据，以及所述申请单与所述结构化医疗数据之间的对应关系。将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述申请单存储到第五数据库中。
在一些实施例中，医院信息系统(HIS)与放射科信息系统(RIS)或成像系统可以通过网络120进行通信。医院信息系统(HIS)获取申请单后，可以通过网络120将申请单发送至放射科信息系统(RIS)或成像系统，放射科信息系统(RIS)或成像系统收到申请单后可以安排拍摄。在一些实施例中，医院信息系统(HIS)与放射科信息系统(RIS)或成像系统之间可能不能通信。放射科信息系统(RIS)或成像系统可以通过对纸质申请单中的文字进行提取及识别，获取申请单中的相关信息。例如，可以通过预设识别算法对申请单中的文本进行提取及识别。在一些实施例中，预设识别算法可以包括光学字符识别算法(OCR，Optical Character Recognition)。光学字符识别算法可以包括但不限于CTC(Connectionist Temporal Classification)算法、CRNN(Convolutional Recurrent Neural Network)算法、CPTN(Connectionist Text Proposal Network)算法、多标签分类(mutli-label classification)算法等。
在一些实施例中，可以先对提取的申请单中的文本进行检错和纠错。例如，可以将提取的申请单中的文本作为待处理文本进行错误检测，得到错误文本的类型和位置。根据错 误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略。根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。关于对提取的申请单中的文本进行检错和纠错处理的更多描述可以参见本说明书其他部分，例如，图12-图22。
步骤2820，进行影像检查，获得目标影像。放射科信息系统(RIS)或成像系统获取申请单后，可以根据申请单安排影像检查，获得目标影像。
步骤2830，基于目标影像生成影像报告。
在一些实施例中，放射科信息系统(RIS)或成像系统生成目标影像后可以根据目标影像生成影像报告。在一些实施例中，放射科信息系统(RIS)或成像系统可以将生成的目标影像发送至PACS和/或后处理工作站，PACS和/或后处理工作站可以根据目标影像生成影像报告。在一些实施例中，可以根据图4-图11的影像报告生成方法自动生成影像报告。
例如，自动生成影像报告的方法可以包括以下步骤：获取目标影像；从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；获取与所述第一特征影像匹配的第二特征影像；获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
在一些实施例中，在根据图4-图11的影像报告生成方法自动生成影像报告时，需要预先构建知识图谱。在预先构建知识图谱的过程中，需要获取原始文本和原始影像。对获取的原始文本可以先进行纠错处理，再进行存储。在一些实施例中，可以将获取的原始文本作为待处理文本，根据图12-图22的文本纠错方法对获取的原始文本进行纠错。从书籍，报刊，网站，电子病历，化验单，诊断报告等中获取的原始文本和原始影像，以及特征影像可以为非结构化数据，可以根据图23-图27的非结构化数据存储方法对原始文本，原始影像，特征影像进行存储。当在特征图像库以及知识图谱中检索特征图像或文本时，可以根据图23-图27所述的方法进行检索。
步骤2840，对影像报告进行检错和纠错。
在一些实施例中，还可以对影像报告中的文本进行错误检测，得到错误文本的类型和位置；根据错误文本类型和纠错策略之间的对应关系，确定与错误文本的类型对应的目标纠错策略；根据目标纠错策略和错误文本的位置，对错误文本进行纠错处理，生成正式报 告。更多描述可以参见本说明书其他部分，例如，图12-图22。
步骤2850，存储影像报告。
在一些实施例中，生成正式报告(非结构化数据)后，还可以对正式报告进行解析，得到正式报告对应的结构化医疗数据，以及正式报告与结构化医疗数据之间的对应关系；将结构化医疗数据和对应关系存储到预设的第四数据库中，将正式报告存储到预设的第五数据库中。
步骤2860，检索影像报告。
在一些实施例中，用户(例如，医护人员)可以通过用户终端130输入查询请求，数据处理系统110可以从第四数据库中查找与输入的查询请求对应的结构化数据，然后根据非结构化数据与结构化数据的对应关系从第五数据库中查找对应的非结构化数据(例如，影像报告)。关于获取、存储、检索非结构化数据的更多描述可以参见本说明书其他部分，例如，图23-图27。
应当注意的是，上述有关流程400的描述仅仅是为了示例和说明，而不限定本说明书的适用范围。对于本领域技术人员来说，在本说明书的指导下可以对流程400进行各种修正和改变。然而，这些修正和改变仍在本说明书的范围之内。
以上概述了成像系统级方法的不同方面和/或通过程序实现其他步骤的方法。技术中的程序部分可以被认为是以可执行的代码和/或相关数据的形式而存在的“产品”或“制品”，是通过计算机可读的介质所参与或实现的。有形的、永久的储存介质包括任何计算机、处理器、或类似设备或相关的模块所用到的内存或存储器。例如各种半导体存储器、磁带驱动器、磁盘驱动器或者类似任何时间能够为软件提供存储功能的设备。
所有软件或其中的一部分有时可能会通过网络进行通信，如互联网或其他通信网络。此类通信能够将软件从一个计算机设备或处理器加载到另一个。例如：从成像系统的一个管理服务器或主机计算机加载至一个计算机环境的硬件平台，或其他实现系统的计算机环境，或与提供按需服务所需要的信息相关的类似功能的系统。因此，另一种能够传递软件元素的介质也可以被用作局部设备之间的物理连接，例如光波、电波、电磁波等，通过电缆、光缆或者空气实现传播。用来载波的物理介质如电缆、无线连接或光缆等类似设备，也可以被认为是承载软件的介质。在这里的用法除非限制了有形的“储存”介质，其他表示计算机或机器“可读介质”的术语都表示在处理器执行任何指令的过程中参与的介质。
因此，一个计算机可读的介质可能有多种形式，包括但不限于，有形的存储介 质，载波介质或物理传输介质。稳定的储存介质包括：光盘或磁盘，以及其他计算机或类似设备中使用的，能够实现图中所描述的系统组件的存储系统。不稳定的存储介质包括动态内存，例如计算机平台的主内存。有形的传输介质包括同轴电缆、铜电缆以及光纤，包括计算机系统内部形成总线的线路。载波传输介质可以传递电信号、电磁信号，声波信号或光波信号，这些信号可以由无线电频率或红外数据通信的方法所产生的。通常的计算机可读介质包括硬盘、软盘、磁带、任何其他磁性介质；CD-ROM、DVD、DVD-ROM、任何其他光学介质；穿孔卡、任何其他包含小孔模式的物理存储介质；RAM、PROM、EPROM、FLASH-EPROM，任何其他存储器片或磁带；传输数据或指令的载波、电缆或传输载波的连接装置、任何其他可以利用计算机读取的程序代码和/或数据。这些计算机可读介质的形式中，会有很多种出现在处理器在执行指令、传递一个或更多结果的过程之中。
本领域技术人员能够理解，本申请所披露的内容可以出现多种变型和改进。例如，以上所描述的不同系统组件都是通过硬件设备所实现的，但是也可能只通过软件的解决方案得以实现。例如：在现有的服务器上安装系统。此外，这里所披露的位置信息的提供可能是通过一个固件、固件/软件的组合、固件/硬件的组合或硬件/固件/软件的组合得以实现。
以上内容描述了本申请和/或一些其他的示例。根据上述内容，本申请还可以作出不同的变形。本申请披露的主题能够以不同的形式和例子所实现，并且本申请可以被应用于大量的应用程序中。后文权利要求中所要求保护的所有应用、修饰以及改变都属于本申请的范围。

Claims (75)

  1. 一种方法，实现于包括至少一个存储设备和至少一个处理设备的装置上，其特征在于，所述方法包括：
    获取目标影像；
    从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；
    获取与所述第一特征影像匹配的第二特征影像；
    获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；
    基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；
    根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
  2. 根据权利要求1所述的方法，其特征在于，构建所述知识图谱包括：
    确定所述知识图谱的图谱结构；
    获取预设领域的原始文本和原始影像；
    从所述原始文本和所述原始影像中提取出多个实体数据；
    确定所述多个实体数据之间的关联关系；
    根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述领域知识图谱。
  3. 根据权利要求2所述的方法，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始文本输入到预先训练的分词模型中，得到所述分词模型输出的多个分词结果；
    将所述多个分词结果输入到预先训练的实体识别模型中，得到所述实体识别模型输出的特征文本，并将所述特征文本确定为所述实体数据。
  4. 根据权利要求3所述的方法，其特征在于，所述确定所述多个实体数据之间的关联关系，包括：
    将所述原始文本和多个所述特征文本输入到预先训练的关系提取模型中，得到所述关系提取模型输出的多个所述特征文本之间的关联关系。
  5. 根据权利要求2-4任一项所述的方法，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始影像输入到预先训练的识别模型中，得到所述识别模型输出的识别结果，所述识别结果用于指示所述原始影像中的感兴趣区域；
    从所述原始影像中截取出所述感兴趣区域的特征影像；
    将所述感兴趣区域的所述特征影像确定为所述实体数据。
  6. 根据权利要求2-5任一项所述的方法，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，确定所述多个实体数据之间的关联关系，包括：
    从所述原始文本中提取出所述原始影像对应的影像描述文本和诊断结果文本；
    从所述原始影像中提取出特征影像；
    将所述影像描述文本，所述诊断结果文本，和所述原始影像的所述特征影像确定为所述实体数据；
    建立所述原始影像的所述特征影像、所述影像描述文本和所述诊断结果文本之间的关联关系。
  7. 根据权利要求2-6任一项所述的方法，其特征在于，所述根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述知识图谱，包括：
    根据所述多个实体数据之间的关联关系将所述多个实体数据组成多个三元组；
    将所述多个三元组填充到所述图谱结构中，得到所述知识图谱。
  8. 根据权利要求1-7任一项所述的方法，其特征在于，所述方法还包括：
    基于所述领域知识图谱和所述第二特征影像，查找与所述第二特征影像对应的疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个；
    根据所述目标影像、所述疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个，生成所述目标影像的影像报告。
  9. 根据权利要求1-8任一项所述的方法，进一步包括：
    对所述目标影像的影像报告进行错误检测，得到错误文本的类型和位置；
    根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  10. 根据权利要求9所述的方法，其特征在于，所述错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本；
    根据所述至少一个候选文本，对所述错误文本进行纠错处理。
  11. 根据权利要求10所述的方法，其特征在于，若所述错误文本的类型包括错字类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述错字类型的文本的位置在所述影像报告中提取出所述错字类型的文本；
    从第一数据库中选取出与所述错字类型的文本同音，和/或，与所述错字类型的文本之间的相似度大于相似度阈值的文本作为所述候选文本。
  12. 根据权利要求10或11所述的方法，其特征在于，若所述错误文本的类型包括缺失类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述缺失类型的文本的位置在所述影像报告中提取出与所述缺失类型的文本相邻的文本；
    从第二数据库中选取出与所述相邻的文本匹配的文本作为所述候选文本。
  13. 根据权利要求10-12任一项所述的方法，其特征在于，若所述错误文本的类型包括无序类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述无序类型的文本的位置在所述影像报告中提取出所述无序类型的文本；
    从第三数据库中选取出与所述无序类型的文本包含相同字的文本作为所述候选文本。
  14. 根据权利要求10-13任一项所述的方法，其特征在于，所述根据所述至少一个候选文本，对所述错误文本进行纠错处理，包括：
    将至少一个所述候选文本依次添加到所述错误文本对应位置上，得到待处理文本；
    对所述待处理文本的通顺程度进行打分；
    按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
  15. 根据权利要求9-14任一项所述的方法，其特征在于，所述错误文本的类型包括冗余类型，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本；
    若确定删除，则删除所述错误文本；
    若确定不删除，则不执行删除操作。
  16. 根据权利要求15所述的方法，其特征在于，所述根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本，包括：
    根据所述冗余类型的文本的位置在所述待影像报告中提取出包含所述冗余类型的文本；
    检测包含所述冗余类型的文本是否通顺；
    若不通顺，则确定删除所述冗余类型的文本；
    若通顺，则确定不删除所述冗余类型的文本。
  17. 根据权利要求1-16任一项所述的方法，其特征在于，所述影像报告为非结构化医疗数据，所述方法进一步包括：
    对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系；
    将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述影像报告存储到第五数据库中。
  18. 根据权利要求17所述的方法，其特征在于，所述对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系，包括：
    根据所述影像报告的非结构化医疗数据类型，从多个预设的解析规则中查找对应得解析规则；
    根据所述对应解析规则，对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据；
    根据所述影像报告的非结构化医疗数据的标识和所述结构化医疗数据的标识，建立所述影像报告和所述结构化医疗数据之间的对应关系。
  19. 根据权利要求18所述的方法，其特征在于，所述多个预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
  20. 根据权利要求17-19任一项所述的方法，其特征在于，所述方法进一步包括：
    通过预设的应用程序接口，接收对应的医疗设备发送的所述影像报告；
    或者，
    根据目标地址标识，在所述目标地址标识对应的医疗设备中进行查找，得到所述影像报告。
  21. 根据权利要求17-20任一项所述的方法，其特征在于，所述方法还包括：
    接收查询请求，所述查询请求携带有所述影像报告的数据标识；
    基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告。
  22. 根据权利要求21所述的方法，其特征在于，所述基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告，包括：
    基于所述查询请求，根据所述数据标识在所述第四数据库中进行查询，得到所述数据标识对应的结构化医疗数据；
    根据所述数据标识对应的结构化医疗数据和所述对应关系，在所述第五数据库中进行查询，得到所述影像报告。
  23. 一种数据处理方法，实现于包括至少一个存储设备和至少一个处理设备的装置上，其特征在于，所述方法包括：
    对待处理文本进行错误检测，得到错误文本的类型和位置；
    根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  24. 一种非结构化数据存储方法，实现于包括至少一个存储设备和至少一个处理设备的装置上，所述方法包括：
    获取待存储的非结构化医疗数据；
    对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；
    将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
  25. 一种系统，包括：
    至少一个存储设备，包括一组指令；
    至少一个处理器，与所述至少一个存储设备通信，其中当执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：
    获取目标影像；
    从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；
    获取与所述第一特征影像匹配的第二特征影像；
    获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；
    基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；
    根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的 影像报告。
  26. 根据权利要求25所述的系统，其特征在于，构建所述知识图谱包括：
    确定所述知识图谱的图谱结构；
    获取预设领域的原始文本和原始影像；
    从所述原始文本和所述原始影像中提取出多个实体数据；
    确定所述多个实体数据之间的关联关系；
    根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述领域知识图谱。
  27. 根据权利要求26所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始文本输入到预先训练的分词模型中，得到所述分词模型输出的多个分词结果；
    将所述多个分词结果输入到预先训练的实体识别模型中，得到所述实体识别模型输出的特征文本，并将所述特征文本确定为所述实体数据。
  28. 根据权利要求27所述的系统，其特征在于，所述确定所述多个实体数据之间的关联关系，包括：
    将所述原始文本和多个所述特征文本输入到预先训练的关系提取模型中，得到所述关系提取模型输出的多个所述特征文本之间的关联关系。
  29. 根据权利要求26-28任一项所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始影像输入到预先训练的识别模型中，得到所述识别模型输出的识别结果，所述识别结果用于指示所述原始影像中的感兴趣区域；
    从所述原始影像中截取出所述感兴趣区域的特征影像；
    将所述感兴趣区域的所述特征影像确定为所述实体数据。
  30. 根据权利要求26-29任一项所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，确定所述多个实体数据之间的关联关系，包括：
    从所述原始文本中提取出所述原始影像对应的影像描述文本和诊断结果文本；
    从所述原始影像中提取出特征影像；
    将所述影像描述文本，所述诊断结果文本，和所述原始影像的所述特征影像确定为所述实体数据；
    建立所述原始影像的所述特征影像、所述影像描述文本和所述诊断结果文本之间的关联关系。
  31. 根据权利要求26-30任一项所述的系统，其特征在于，所述根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述知识图谱，包括：
    根据所述多个实体数据之间的关联关系将所述多个实体数据组成多个三元组；
    将所述多个三元组填充到所述图谱结构中，得到所述知识图谱。
  32. 根据权利要求25-31任一项所述的系统，其特征在于，所述至少一个处理器被配置为使所述系统执行所述至少一个操作，包括：
    基于所述领域知识图谱和所述第二特征影像，查找与所述第二特征影像对应的疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个；
    根据所述目标影像、所述疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个，生成所述目标影像的影像报告。
  33. 根据权利要求25-32任一项所述的系统，其特征在于，所述至少一个处理器被配置为使所述系统执行所述至少一个操作，包括：
    对所述目标影像的影像报告进行错误检测，得到错误文本的类型和位置；
    根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  34. 根据权利要求33所述的系统，其特征在于，所述错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候 选文本；
    根据所述至少一个候选文本，对所述错误文本进行纠错处理。
  35. 根据权利要求34所述的系统，其特征在于，若所述错误文本的类型包括错字类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述错字类型的文本的位置在所述影像报告中提取出所述错字类型的文本；
    从第一数据库中选取出与所述错字类型的文本同音，和/或，与所述错字类型的文本之间的相似度大于相似度阈值的文本作为所述候选文本。
  36. 根据权利要求34或35所述的系统，其特征在于，若所述错误文本的类型包括缺失类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述缺失类型的文本的位置在所述影像报告中提取出与所述缺失类型的文本相邻的文本；
    从第二数据库中选取出与所述相邻的文本匹配的文本作为所述候选文本。
  37. 根据权利要求34-36任一项所述的系统，其特征在于，若所述错误文本的类型包括无序类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述无序类型的文本的位置在所述影像报告中提取出所述无序类型的文本；
    从第三数据库中选取出与所述无序类型的文本包含相同字的文本作为所述候选文本。
  38. 根据权利要求34-37任一项所述的系统，其特征在于，所述根据所述至少一个候选文本，对所述错误文本进行纠错处理，包括：
    将至少一个所述候选文本依次添加到所述错误文本对应位置上，得到待处理文本；
    对所述待处理文本的通顺程度进行打分；
    按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
  39. 根据权利要求33-38任一项所述的系统，其特征在于，所述错误文本的类型包括冗余类型，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本；
    若确定删除，则删除所述错误文本；
    若确定不删除，则不执行删除操作。
  40. 根据权利要求39所述的系统，其特征在于，所述根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本，包括：
    根据所述冗余类型的文本的位置在所述待影像报告中提取出包含所述冗余类型的文本；
    检测包含所述冗余类型的文本是否通顺；
    若不通顺，则确定删除所述冗余类型的文本；
    若通顺，则确定不删除所述冗余类型的文本。
  41. 根据权利要求25-40任一项所述的系统，其特征在于，所述影像报告为非结构化医疗数据，所述至少一个处理器被配置为使所述系统执行所述至少一个操作，包括：
    对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系；
    将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述影像报告存储到第五数据库中。
  42. 根据权利要求41所述的系统，其特征在于，所述对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系，包括：
    根据所述影像报告的非结构化医疗数据类型，从多个预设的解析规则中查找对应得解析规则；
    根据所述对应解析规则，对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据；
    根据所述影像报告的非结构化医疗数据的标识和所述结构化医疗数据的标识，建立所述影像报告和所述结构化医疗数据之间的对应关系。
  43. 根据权利要求42所述的系统，其特征在于，所述多个预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
  44. 根据权利要求41-43任一项所述的系统，其特征在于，所述至少一个处理器被配置为使所述系统执行所述至少一个操作，包括：
    通过预设的应用程序接口，接收对应的医疗设备发送的所述影像报告；或者
    根据目标地址标识，在所述目标地址标识对应的医疗设备中进行查找，得到所述影像报告。
  45. 根据权利要求41-44任一项所述的系统，其特征在于，所述至少一个处理器被配置为使所述系统执行所述至少一个操作，包括：
    接收查询请求，所述查询请求携带有所述影像报告的数据标识；
    基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告。
  46. 根据权利要求45所述的系统，其特征在于，所述基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告，包括：
    基于所述查询请求，根据所述数据标识在所述第四数据库中进行查询，得到所述数据标识对应的结构化医疗数据；
    根据所述数据标识对应的结构化医疗数据和所述对应关系，在所述第五数据库中进行查询，得到所述影像报告。
  47. 一种数据处理系统，包括：
    至少一个存储设备，包括一组指令；
    至少一个处理器，与所述至少一个存储设备通信，其中当执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：
    对待处理文本进行错误检测，得到错误文本的类型和位置；
    根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应 的目标纠错策略；
    根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  48. 一种非结构化数据存储系统，包括：
    至少一个存储设备，包括一组指令；
    至少一个处理器，与所述至少一个存储设备通信，其中当执行该组指令时，所述至少一个处理器被配置为使所述系统执行至少一个操作，包括：
    获取待存储的非结构化医疗数据；
    对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；
    将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
  49. 一种系统，包括：
    影像截取模块，被配置为
    获取目标影像；以及
    从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；
    影像查找模块，被配置为获取与所述第一特征影像匹配的第二特征影像；
    信息查找模块，被配置为
    获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；以及
    基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；以及
    报告生成模块，被配置为根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
  50. 根据权利要求49所述的系统，其特征在于，构建所述知识图谱包括：
    确定所述知识图谱的图谱结构；
    获取预设领域的原始文本和原始影像；
    从所述原始文本和所述原始影像中提取出多个实体数据；
    确定所述多个实体数据之间的关联关系；
    根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述领域知识图谱。
  51. 根据权利要求50所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始文本输入到预先训练的分词模型中，得到所述分词模型输出的多个分词结果；
    将所述多个分词结果输入到预先训练的实体识别模型中，得到所述实体识别模型输出的特征文本，并将所述特征文本确定为所述实体数据。
  52. 根据权利要求51所述的系统，其特征在于，所述确定所述多个实体数据之间的关联关系，包括：
    将所述原始文本和多个所述特征文本输入到预先训练的关系提取模型中，得到所述关系提取模型输出的多个所述特征文本之间的关联关系。
  53. 根据权利要求50-52任一项所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，包括：
    将所述原始影像输入到预先训练的识别模型中，得到所述识别模型输出的识别结果，所述识别结果用于指示所述原始影像中的感兴趣区域；
    从所述原始影像中截取出所述感兴趣区域的特征影像；
    将所述感兴趣区域的所述特征影像确定为所述实体数据。
  54. 根据权利要求50-53任一项所述的系统，其特征在于，所述从所述原始文本和所述原始影像中提取出多个实体数据，确定所述多个实体数据之间的关联关系，包括：
    从所述原始文本中提取出所述原始影像对应的影像描述文本和诊断结果文本；
    从所述原始影像中提取出特征影像；
    将所述影像描述文本，所述诊断结果文本，和所述原始影像的所述特征影像确定为所述实体数据；
    建立所述原始影像的所述特征影像、所述影像描述文本和所述诊断结果文本之间的关联 关系。
  55. 根据权利要求50-54任一项所述的系统，其特征在于，所述根据所述图谱结构、所述多个实体数据和所述多个实体数据之间的关联关系构建所述知识图谱，包括：
    根据所述多个实体数据之间的关联关系将所述多个实体数据组成多个三元组；
    将所述多个三元组填充到所述图谱结构中，得到所述知识图谱。
  56. 根据权利要求49-55任一项所述的系统，其特征在于，
    所述信息查找模块被配置为基于所述领域知识图谱和所述第二特征影像，查找与所述第二特征影像对应的疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个；
    所述报告生成模块被配置为根据所述目标影像、所述疾病信息、诱因信息、症状信息、药物信息或治疗方案信息中的至少一个，生成所述目标影像的影像报告。
  57. 根据权利要求49-56任一项所述的系统，进一步包括：
    检测模块，被配置为对所述目标影像的影像报告进行错误检测，得到错误文本的类型和位置；
    确定模块，被配置为根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    纠错模块，被配置为根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  58. 根据权利要求57所述的系统，其特征在于，所述错误文本的类型包括错字类型、缺失类型、无序类型中的至少一种，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本；
    根据所述至少一个候选文本，对所述错误文本进行纠错处理。
  59. 根据权利要求58所述的系统，其特征在于，若所述错误文本的类型包括错字类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个 候选文本，包括：
    根据所述错字类型的文本的位置在所述影像报告中提取出所述错字类型的文本；
    从第一数据库中选取出与所述错字类型的文本同音，和/或，与所述错字类型的文本之间的相似度大于相似度阈值的文本作为所述候选文本。
  60. 根据权利要求58或59所述的系统，其特征在于，若所述错误文本的类型包括缺失类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述缺失类型的文本的位置在所述影像报告中提取出与所述缺失类型的文本相邻的文本；
    从第二数据库中选取出与所述相邻的文本匹配的文本作为所述候选文本。
  61. 根据权利要求58-60任一项所述的系统，其特征在于，若所述错误文本的类型包括无序类型，则所述根据所述目标纠错策略和所述错误文本的位置，确定与所述错误文本对应的至少一个候选文本，包括：
    根据所述无序类型的文本的位置在所述影像报告中提取出所述无序类型的文本；
    从第三数据库中选取出与所述无序类型的文本包含相同字的文本作为所述候选文本。
  62. 根据权利要求58-61任一项所述的系统，其特征在于，所述根据所述至少一个候选文本，对所述错误文本进行纠错处理，包括：
    将至少一个所述候选文本依次添加到所述错误文本对应位置上，得到待处理文本；
    对所述待处理文本的通顺程度进行打分；
    按照分数从高到低的顺序对多个打分后的待处理文本进行排序，并将分数最高的待处理文本确定为纠正后的文本。
  63. 根据权利要求57-62任一项所述的系统，其特征在于，所述错误文本的类型包括冗余类型，则所述根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理，包括：
    根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本；
    若确定删除，则删除所述错误文本；
    若确定不删除，则不执行删除操作。
  64. 根据权利要求63所述的系统，其特征在于，所述根据所述目标纠错策略和所述错误文本的位置，确定是否删除所述冗余类型的文本，包括：
    根据所述冗余类型的文本的位置在所述待影像报告中提取出包含所述冗余类型的文本；
    检测包含所述冗余类型的文本是否通顺；
    若不通顺，则确定删除所述冗余类型的文本；
    若通顺，则确定不删除所述冗余类型的文本。
  65. 根据权利要求49-64任一项所述的系统，其特征在于，所述影像报告为非结构化医疗数据，所述系统进一步包括：
    解析模块，被配置为对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系；
    存储模块，被配置为将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述影像报告存储到第五数据库中。
  66. 根据权利要求65所述的系统，其特征在于，所述对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据，以及所述影像报告与所述结构化医疗数据之间的对应关系，包括：
    根据所述影像报告的非结构化医疗数据类型，从多个预设的解析规则中查找对应得解析规则；
    根据所述对应解析规则，对所述影像报告进行解析，得到所述影像报告对应的结构化医疗数据；
    根据所述影像报告的非结构化医疗数据的标识和所述结构化医疗数据的标识，建立所述影像报告和所述结构化医疗数据之间的对应关系。
  67. 根据权利要求66所述的系统，其特征在于，所述多个预设的解析规则包括：医学影像文字识别规则、DICOM文件解析规则和XML报告解析规则。
  68. 根据权利要求65-67任一项所述的系统，进一步包括获取模块，被配置为：
    通过预设的应用程序接口，接收对应的医疗设备发送的所述影像报告；或者
    根据目标地址标识，在所述目标地址标识对应的医疗设备中进行查找，得到所述影像报告。
  69. 根据权利要求65-68任一项所述的系统，进一步包括：
    接收模块，被配置为接收查询请求，所述查询请求携带有所述影像报告的数据标识；
    查询模块，被配置为基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告。
  70. 根据权利要求69所述的系统，其特征在于，所述基于所述查询请求，根据所述数据标识和所述对应关系，分别在所述第四数据库和所述第五数据库进行查询，得到所述数据标识对应的结构化医疗数据和所述影像报告，包括：
    基于所述查询请求，根据所述数据标识在所述第四数据库中进行查询，得到所述数据标识对应的结构化医疗数据；
    根据所述数据标识对应的结构化医疗数据和所述对应关系，在所述第五数据库中进行查询，得到所述影像报告。
  71. 一种数据处理系统，包括：
    检测模块，被配置为对待处理文本进行错误检测，得到错误文本的类型和位置；
    确定模块，被配置为根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    纠错模块，被配置为根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  72. 一种非结构化数据存储系统，包括：
    获取模块，被配置为获取待存储的非结构化医疗数据；
    解析模块，被配置为对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；
    存储模块，被配置为将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
  73. 一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：
    获取目标影像；
    从所述目标影像中截取出第一特征影像，其中，所述第一特征影像包括所述目标影像中感兴趣区域的影像；
    获取与所述第一特征影像匹配的第二特征影像；
    获取知识图谱，所述知识图谱包括第二特征影像与对应的影像描述信息、诊断结果信息的关联关系；
    基于所述知识图谱和所述第二特征影像，查找出与所述第二特征影像对应的影像描述信息和诊断结果信息；
    根据所述目标影像、所述影像描述信息和所述诊断结果信息生成所述目标影像的影像报告。
  74. 一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：
    对待处理文本进行错误检测，得到错误文本的类型和位置；
    根据错误文本类型和纠错策略之间的对应关系，确定与所述错误文本的类型对应的目标纠错策略；
    根据所述目标纠错策略和所述错误文本的位置，对所述错误文本进行纠错处理。
  75. 一种非暂时性计算机可读介质，包括至少一组指令集，其中所述至少一组指令集由计算设备的一个或以上处理器执行时，所述至少一组指令集使得所述计算设备执行一种方法，所述方法包括：
    获取待存储的非结构化医疗数据；
    对所述非结构化医疗数据进行解析，得到所述非结构化医疗数据对应的结构化医疗数据，以及所述非结构化医疗数据与所述结构化医疗数据之间的对应关系；
    将所述结构化医疗数据和所述对应关系存储到第四数据库中，将所述非结构化医疗数据存储到第五数据库中。
PCT/CN2021/106971 2020-07-17 2021-07-17 医疗数据处理方法与系统 Ceased WO2022012687A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP21843125.2A EP4170670A4 (en) 2020-07-17 2021-07-17 Medical data processing method and system
US18/155,710 US12562257B2 (en) 2020-07-17 2023-01-17 Systems and methods for medical data processing

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
CN202010690317.8A CN111950262A (zh) 2020-07-17 2020-07-17 数据处理方法、装置、计算机设备和存储介质
CN202010690317.8 2020-07-17
CN202011434314.4A CN112530550A (zh) 2020-12-10 2020-12-10 影像报告生成方法、装置、计算机设备和存储介质
CN202011434314.4 2020-12-10
CN202011495360.5 2020-12-17
CN202011495360.5A CN112527923A (zh) 2020-12-17 2020-12-17 非结构化数据存储方法、装置、计算机设备和存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/155,710 Continuation US12562257B2 (en) 2020-07-17 2023-01-17 Systems and methods for medical data processing

Publications (1)

Publication Number Publication Date
WO2022012687A1 true WO2022012687A1 (zh) 2022-01-20

Family

ID=79554498

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/106971 Ceased WO2022012687A1 (zh) 2020-07-17 2021-07-17 医疗数据处理方法与系统

Country Status (3)

Country Link
US (1) US12562257B2 (zh)
EP (1) EP4170670A4 (zh)
WO (1) WO2022012687A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114328938A (zh) * 2022-03-16 2022-04-12 浙江卡易智慧医疗科技有限公司 一种影像报告结构化提取方法
CN114724670A (zh) * 2022-06-02 2022-07-08 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) 一种医学报告生成方法、装置、存储介质和电子设备
CN115631871A (zh) * 2022-12-22 2023-01-20 北京大学第三医院（北京大学第三临床医学院） 一种药物相互作用等级的确定方法和装置

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US12210827B2 (en) * 2021-08-23 2025-01-28 International Business Machines Corporation Specificity ranking of text elements and applications thereof
US12561524B2 (en) * 2023-08-24 2026-02-24 International Business Machines Corporation Training machine learning models to automatically detect and correct contextual and logical errors
CN117152780A (zh) * 2023-09-01 2023-12-01 北京京东拓先科技有限公司 药物图片的处理方法、装置、电子设备和存储介质
CN117238440A (zh) * 2023-09-07 2023-12-15 浙江远图技术股份有限公司 智慧药品配送系统及方法
US12597504B2 (en) * 2023-11-17 2026-04-07 Siemens Healthineers Ag Automatic selection and display layout of medical images from clinical descriptions
CN118823790B (zh) * 2024-09-18 2024-12-17 北京联影智能影像技术研究院 医学影像和影像报告的交互方法、装置、计算机设备、存储介质和程序产品

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583440A (zh) * 2017-09-28 2019-04-05 北京西格码列顿信息技术有限公司 结合影像识别与报告编辑的医学影像辅助诊断方法及系统
CN110704413A (zh) * 2019-09-30 2020-01-17 东莞市盟大塑化科技有限公司 一种基于深度学习的知识图谱构建方法
CN110911009A (zh) * 2019-11-14 2020-03-24 南京医科大学 一种临床诊断辅助决策系统和医学知识图谱积累方法
CN111090986A (zh) * 2019-11-29 2020-05-01 福建亿榕信息技术有限公司 一种公文文档纠错的方法
CN111223076A (zh) * 2019-12-30 2020-06-02 北京深睿博联科技有限责任公司 神经系统影像分析方法和装置
CN111274785A (zh) * 2020-01-21 2020-06-12 北京字节跳动网络技术有限公司 一种文本纠错方法、装置、设备及介质
CN111292821A (zh) * 2020-01-21 2020-06-16 上海联影智能医疗科技有限公司 一种医学诊疗系统
CN111324905A (zh) * 2020-02-17 2020-06-23 平安国际智慧城市科技股份有限公司 影像数据标注方法、装置、计算机设备和存储介质
CN111369996A (zh) * 2020-02-24 2020-07-03 网经科技（苏州）有限公司 一种特定领域的语音识别文本纠错方法
CN111950262A (zh) * 2020-07-17 2020-11-17 武汉联影医疗科技有限公司 数据处理方法、装置、计算机设备和存储介质
CN112527923A (zh) * 2020-12-17 2021-03-19 武汉联影医疗科技有限公司 非结构化数据存储方法、装置、计算机设备和存储介质
CN112530550A (zh) * 2020-12-10 2021-03-19 武汉联影医疗科技有限公司 影像报告生成方法、装置、计算机设备和存储介质

Family Cites Families (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11048874B2 (en) * 2016-01-05 2021-06-29 International Business Machines Corporation Medical record error detection system and method
CN107301241B (zh) 2017-07-03 2020-06-05 北京奇艺世纪科技有限公司 一种视频搜索中的查询词纠错方法和装置
CN110019839B (zh) * 2018-01-03 2021-11-05 中国科学院计算技术研究所 基于神经网络和远程监督的医学知识图谱构建方法和系统
CN109378053B (zh) * 2018-11-30 2021-07-06 安徽影联云享医疗科技有限公司 一种用于医学影像的知识图谱构建方法
CN109800414B (zh) 2018-12-13 2023-04-18 科大讯飞股份有限公司 语病修正推荐方法及系统
CN110010219B (zh) 2019-03-13 2021-12-10 杭州电子科技大学 光学相干层析图像视网膜病变智能检测系统及检测方法
CN110111905B (zh) 2019-04-24 2021-09-03 云知声智能科技股份有限公司 一种医疗知识图谱的构建系统和构建方法
CN110210025A (zh) 2019-05-29 2019-09-06 广州伟宏智能科技有限公司 一种基于文本提取的转换方法
CN110335256A (zh) 2019-06-18 2019-10-15 广州智睿医疗科技有限公司 一种病理辅助诊断方法
CN110457688B (zh) 2019-07-23 2023-11-24 广州视源电子科技股份有限公司 纠错处理方法及装置、存储介质和处理器
CN110555153A (zh) 2019-08-20 2019-12-10 暨南大学 一种基于领域知识图谱的问答系统及其构建方法
CN110458373A (zh) 2019-08-22 2019-11-15 电子科技大学成都学院 一种基于知识图谱融合的犯罪预测方法及系统
CN110570928A (zh) 2019-09-11 2019-12-13 浪潮软件股份有限公司 一种基于HBase和ozone的医疗影像文件存取方法
CN110827990B (zh) 2019-10-31 2022-08-19 北京科技大学 一种基于知识图谱的伤寒辨证推理系统
CN111241212B (zh) * 2020-01-20 2023-10-24 京东方科技集团股份有限公司 知识图谱的构建方法及装置、存储介质、电子设备
CN111339758B (zh) 2020-02-21 2023-06-30 苏宁云计算有限公司 基于深度学习模型的文本纠错方法及系统
CN111368094B (zh) 2020-02-27 2024-03-26 沈阳东软熙康医疗系统有限公司 实体知识图谱建立、属性信息获取、门诊分诊方法及装置
CN111401012B (zh) 2020-03-09 2023-11-21 北京声智科技有限公司 文本纠错方法、电子设备及计算机可读存储介质
CN111310447B (zh) 2020-03-18 2024-02-02 河北省讯飞人工智能研究院 语法纠错方法、装置、电子设备和存储介质
CN111640480B (zh) 2020-05-21 2023-09-26 上海联影智能医疗科技有限公司 医学报告生成方法、计算机设备和存储介质
CN111767410B (zh) 2020-06-30 2023-05-30 深圳平安智慧医健科技有限公司 临床医疗知识图谱的构建方法、装置、设备及存储介质
CN111950282A (zh) 2020-07-08 2020-11-17 国家计算机网络与信息安全管理中心 一种基于网络行为特征的扩线分析方法及装置
CN111951952A (zh) 2020-07-17 2020-11-17 北京欧应信息技术有限公司 一种基于医疗影像信息自动诊断骨科疾病的装置
CN112035404B (zh) 2020-08-28 2023-02-10 康键信息技术（深圳）有限公司 医疗数据监控与预警方法、装置、设备及存储介质

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583440A (zh) * 2017-09-28 2019-04-05 北京西格码列顿信息技术有限公司 结合影像识别与报告编辑的医学影像辅助诊断方法及系统
CN110704413A (zh) * 2019-09-30 2020-01-17 东莞市盟大塑化科技有限公司 一种基于深度学习的知识图谱构建方法
CN110911009A (zh) * 2019-11-14 2020-03-24 南京医科大学 一种临床诊断辅助决策系统和医学知识图谱积累方法
CN111090986A (zh) * 2019-11-29 2020-05-01 福建亿榕信息技术有限公司 一种公文文档纠错的方法
CN111223076A (zh) * 2019-12-30 2020-06-02 北京深睿博联科技有限责任公司 神经系统影像分析方法和装置
CN111274785A (zh) * 2020-01-21 2020-06-12 北京字节跳动网络技术有限公司 一种文本纠错方法、装置、设备及介质
CN111292821A (zh) * 2020-01-21 2020-06-16 上海联影智能医疗科技有限公司 一种医学诊疗系统
CN111324905A (zh) * 2020-02-17 2020-06-23 平安国际智慧城市科技股份有限公司 影像数据标注方法、装置、计算机设备和存储介质
CN111369996A (zh) * 2020-02-24 2020-07-03 网经科技（苏州）有限公司 一种特定领域的语音识别文本纠错方法
CN111950262A (zh) * 2020-07-17 2020-11-17 武汉联影医疗科技有限公司 数据处理方法、装置、计算机设备和存储介质
CN112530550A (zh) * 2020-12-10 2021-03-19 武汉联影医疗科技有限公司 影像报告生成方法、装置、计算机设备和存储介质
CN112527923A (zh) * 2020-12-17 2021-03-19 武汉联影医疗科技有限公司 非结构化数据存储方法、装置、计算机设备和存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4170670A4

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114328938A (zh) * 2022-03-16 2022-04-12 浙江卡易智慧医疗科技有限公司 一种影像报告结构化提取方法
CN114724670A (zh) * 2022-06-02 2022-07-08 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) 一种医学报告生成方法、装置、存储介质和电子设备
CN115631871A (zh) * 2022-12-22 2023-01-20 北京大学第三医院（北京大学第三临床医学院） 一种药物相互作用等级的确定方法和装置

Also Published As

Publication number Publication date
EP4170670A4 (en) 2023-12-27
US12562257B2 (en) 2026-02-24
EP4170670A1 (en) 2023-04-26
US20230154593A1 (en) 2023-05-18

Similar Documents

Publication Publication Date Title
WO2022012687A1 (zh) 医疗数据处理方法与系统
Porturas et al. Forty years of emergency medicine research: Uncovering research themes and trends through topic modeling
CN108920453B (zh) 数据处理方法、装置、电子设备及计算机可读介质
US20220044812A1 (en) Automated generation of structured patient data record
CN113345577B (zh) 诊疗辅助信息的生成方法、模型训练方法、装置、设备以及存储介质
CN112331298B (zh) 开具药方的方法、装置、电子设备及存储介质
CN116992839B (zh) 病案首页自动生成方法、装置及设备
CN111627512A (zh) 一种相似病历的推荐方法、装置，电子设备及存储介质
CN107403067A (zh) 基于医学知识库的智能分诊服务器、终端及系统
US20240006039A1 (en) Medical structured reporting workflow assisted by natural language processing techniques
CN112530550A (zh) 影像报告生成方法、装置、计算机设备和存储介质
WO2022160614A1 (zh) 构建医学实体关系图的方法及装置、医嘱质控的方法及装置、设备、介质
CN112528660A (zh) 处理文本的方法、装置、设备、存储介质和程序产品
CN112883157A (zh) 一种多源异构医疗数据的标准化方法及装置
JP2023514023A (ja) 質問の検索装置、質問の検索方法、デバイス、および記憶媒体
US20250078987A1 (en) Image interpretation model development
CN109346171A (zh) 一种辅助诊断方法、装置和计算机设备
CN112115240A (zh) 分类处理方法、装置、服务器和存储介质
US20250378070A1 (en) Answer generation using machine reading comprehension and supported decision trees
CN116258136A (zh) 检错模型训练方法、医学影像报告检测方法、系统及设备
US20230046367A1 (en) Systems and methods for dynamically removing text from documents
Nair et al. Automated clinical concept-value pair extraction from discharge summary of pituitary adenoma patients
CN120748600A (zh) 基于生殖医疗数据模型的跨系统数据整合方法及装置
RU2701702C2 (ru) Система и способ единообразного сопоставления неструктурированных записанных признаков сo связанными терапевтическими признаками
CN113221541A (zh) 一种数据提取方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21843125

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021843125

Country of ref document: EP

Effective date: 20230118

NENP Non-entry into the national phase

Ref country code: DE