WO2024006586A2 - Système de diagnostic sans contact - Google Patents

Système de diagnostic sans contact Download PDF

Info

Publication number
WO2024006586A2
WO2024006586A2 PCT/US2023/031824 US2023031824W WO2024006586A2 WO 2024006586 A2 WO2024006586 A2 WO 2024006586A2 US 2023031824 W US2023031824 W US 2023031824W WO 2024006586 A2 WO2024006586 A2 WO 2024006586A2
Authority
WO
WIPO (PCT)
Prior art keywords
images
imaging sensor
sensor
sensors
processor
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/US2023/031824
Other languages
English (en)
Other versions
WO2024006586A3 (fr
Inventor
Daniel G. O'connell
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.)
Hnu Photonics LLC
Original Assignee
Hnu Photonics LLC
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
Application filed by Hnu Photonics LLC filed Critical Hnu Photonics LLC
Publication of WO2024006586A2 publication Critical patent/WO2024006586A2/fr
Publication of WO2024006586A3 publication Critical patent/WO2024006586A3/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • 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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A61B5/0077—Devices for viewing the surface of the body, e.g. camera, magnifying lens
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/01—Measuring temperature of body parts ; Diagnostic temperature sensing, e.g. for malignant or inflamed tissue
    • A61B5/015—By temperature mapping of body part
    • 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

Definitions

  • the system is used to make health assessments of human subjects and is comprised of a sensor head unit 10, a positioning system 20 and a connected computer 30 ( Figure 1) that may include an internet connection for remote analysis.
  • the positioning system 20 is used to adjust the sensor head unit to the height of the subject, for example using a mechanized elevator 23.
  • the positioning system is utilized to view the subject in the desired direction and orientation to collect data of interest pertaining to a particular health condition.
  • a gimbal 22 can provide the head unit 10 pan 24 and tilt 25control to orient the sensors correctly.
  • the positioning system is controlled by a computer 30 using feature detection algorithms.
  • an artificial intelligence face, mouth, or eye detector can be used to align the sensor head unit based on the location of the region of interest such as face, mouth, or eyes within the field of view (FOV) of the desired sensor.
  • FOV field of view
  • the sensor head unit 10 incorporates a thermal imaging sensor 11, at least one visible or infrared imaging sensor 12 capable of narrow FOV, at least one visible or infrared imaging sensor capable of wide FOV 13 or an imaging sensor capable of both wide and narrow FOV using a mechanized zoom lens, and at least one hyperspectral imaging sensor 14 shown in Error! Reference source not found..
  • the sensor head unit 10 may also include at least one light source 15 attached internal or external to the sensor head for controlled illumination, a LIDAR ("Light Detection and Ranging") sensor 16 or multiple LIDAR sensors 17 , and at least one control circuit board.
  • the LIDAR unit 16 or multiple units 17 may also be used to measure distance or depth information and generate a 3D model of a scene.
  • the LIDAR may be used to ensure the subject is located at a predetermined distance from the head unit as shown in Error! Reference source not found..
  • Figure 1 schematic diagram of a system described herein.
  • Figure 2 schematic diagram of the sensor head unit.
  • Figure 3 diagram showing human subject positioned for multi-sensor disease detection.
  • Figure 4 representations of recorded image data.
  • Figure 5 flowchart showing the process for generating a health assessment score.
  • Figure 6 narrow FOV image of an eye (top) and the same image with an iris mask and sclera mask (bottom).
  • Figure 7 flowchart showing steps in processing of eye images.
  • Figure 8 flowchart showing steps in feature extraction from processed eye images.
  • Figure 9 example of high density histogram (HDH) showing one-dimensional (left) and three-dimensional (right) representations.
  • Figure 10 schematic representation of separation metric is used to identify the evaluation criteria conditions where positive subjects and negative subjects are most distinctive.
  • Figure 11 A recorded picture of a scene consisting of an open mouth (left) and a representation of the collected spectral data (right).
  • Figure 12 A diagram illustrating back projection techniques used to isolate the specific portions of the recorded spectral information that relate to the spectral composition of the target region of interest and produce spectral intensity vs wavelength profiles as an output.
  • Figure 14 Flowchart showing process for extraction of multiple features from spectral, thermal, and imaging sensors.
  • Figure 16 schematic of pre-processing system.
  • Figure 17 schematic of inference process.
  • the system performs disease detection by fusing information from multiple sensors: thermal imaging sensor 11, one or more visible or infrared imaging sensors (12, 13), and one or more hyperspectral sensor. Each sensor produces one or more images or data sets per subject (a person undergoing a test). Error! Reference source not found, shows how information is fused to develop a health assessment score for a particular vital, health feature or disease that the system had been trained on.
  • a set of mouth, face, and eye images are collected for disease positive subjects and a set for disease negative subjects.
  • the system can be trained for a variety of diseases using the same image data set.
  • One sensor 12 of the system utilizes narrow FOV images of the eyes (top of Figure 6) to detect specific diseases through a machine learning training process using the blood vessels and other regions of interest within the eyes.
  • the process was designed so that a minimal number of training subjects would be necessary as it can be time-consuming and expensive to collect the hundreds or thousands of subjects required by currently available a rtificia I intelligence neural network frameworks.
  • the process was further designed for transparency so that decisions made could be interpreted by humans that could then impose knowledge-based algorithm adjustments based on measured pathology features to improve accuracy.
  • artificial neural networks are more like black boxes because the trained "weights" are not humanly interpretable.
  • a spatial segmentation algorithm shown in Block 10 is used to segment the blood vessel mask into individual blood vessels that are spatially isolated from others (i.e., mask of each detected blood vessel is not in contact with others). Each blood vessel is a region of interest. (10) The regions surrounding the individual blood vessels are also of interest. To create these masks, each blood vessel binary mask is dilated. Then the blood vessel mask is subtracted forming a surrounding mask for each blood vessel. These steps are illustrated in Error! Reference source not found, in Blocks 11 and 12. The thickness of these surrounding regions is tunable.
  • the individual blood vessels are sorted over a variety of criteria including but not limited to length, distance from iris, chromatic brightness, chromatic ratios, texture, and any mixture as shown in Error! Reference source not found..
  • Features are extracted from the sorted list. These could include, for example, the ratio of the mean color values inside of a blood vessel with the mean of the color values in the region surrounding the blood vessel. This ratio is compared with the N largest blood vessels. This analysis results in one of several features that will be used in the training process. The sorting and feature extraction are summarized in Error!
  • HDH high dimensional histograms
  • Basic histograms are one dimensional and discretized into a number of bins that contain a count of occurrences of particular quantities within a measurement (such as an image or other sensor output). Histograms are an implementation of Probability Distribution Functions (PDF).
  • PDF Probability Distribution Functions
  • values of every pixel color values are transformed into an HDH.
  • the axes of the HDH are defined by the anatomically rooted features from the images.
  • axis one is the ratio of the red color pixel value to the blue color pixel value
  • axis two is the standard deviation of the green color pixel value over all pixels adjacent to the center pixel
  • axis three is the pixel distance from the iris (Error! Reference source not found.).
  • a separation metric is used to identify the evaluation criteria conditions where positive subjects and negative subjects are most distinctive which are denoted as "extremes" in Error! Reference source not found..
  • the separation metric measures the distance between the average values for the positives and the average values for the negatives, divided by the larger of the two standard deviations.
  • the identified evaluation criteria are used in the training process and from this point forward, the values for these specific HDH axis locations will be treated as "features" as defined above.
  • spectral sensor image information In addition to the narrow FOV images of the eye, Error! Reference source not found, also refers to the use of spectral sensor image information to include in the health assessment process conducted by the system.
  • These spectral sensor images of the subject can include but are not limited to spectral images of the mouth and eyes.
  • An example of this process when collecting spectral image information from an open human mouth is shown in Error! Reference source not found..
  • a recorded picture of a scene consisting of an open mouth is shown on the left portion of the image and a representation of the collected spectral data is shown to the right.
  • Back projection techniques are used to isolate the specific portions of the recorded spectral information that relate to the spectral composition of the target region of interest and produce spectral intensity vs wavelength profiles as an output. This output is used to contribute to the health assessment process conducted by the system.
  • the black region in the mouth indicates the location of the automatically segmented region of interest from the recorded image of the scene. Based on the specified region of interest, the associated portions of the spectral information that relate to the composition of wavelength spectral information for points within the scene ROI are isolated and are analyzed to produce spectral intensity profiles for the given region of interest.
  • the final sensor input referred to in Error! Reference source not found is the thermal imaging sensor, capable of recording images of the subject that includes but is not limited to thermal images of the face.
  • Error! Reference source not found shows a face automatically segmented for the forehead, eyes, nose, cheeks, and mouth. Features from these segmented regions are extracted. These features can be averages, variances, skewness, kurtosis, ratios, and other properties.
  • Another source of features for the disease detection process are spatial gradient profiles. In each of the regions (forehead, eyes, nose, cheeks, and mouth), spatial gradients are observed. As an example, coefficients from polynomial regression algorithms fitted to the spatial gradient profiles are treated as features that will be used in the disease and health assessment process.
  • the multiple sensor dataset using feature extraction used for disease detection as described in this embodiment can also be used to measure human vital signs.
  • the disease detection process outlined in this section is one example of many (including artificial neural networks) that simultaneous uses the features extracted from multiple independent sensors in a training process to maximize disease detection capability.
  • multiple features are extracted from each of the sensors. These features are organized as shown in Error! Reference source not found..
  • features are extracted from the spectral sensor image for both disease negative and positive subjects. These features can be items such as the inflection wavelengths, peaks within the Fourier transform, local peaks at various wavelengths, or other features.
  • thermal images statistical and other features are extracted along with coefficients from fitting directional profiles within the regions of interest.
  • features are extracted along with the HDH's for both disease negative and positive subjects.
  • the eye image HDH from the disease negative subjects are compared with those from the disease positive subjects. This comparison looks for HDH locations that maximally separate disease positives from negatives as previously shown in Error! Reference source not found..
  • the feature extraction step results in a large number of features for each disease positive and negative subject. These features individually might have limited accuracy in their ability to identify disease positive and negative subjects, but when multiple features are used jointly, the accuracy will be increased provided a good selection of highly separative independent features. This jointness is achieved by organizing the features into dimensions. This organization behaves like a classical Bayesian classifier, but for several simultaneous known features. In Error! Reference source not found., the "High Sep. Bins" block identifies the extremes from Error! Reference source not found..
  • the classifier uses a limited number of dimensions, D, which is far less than the number of K total features extracted. For example, D may range from 3 to 9 dimensions or higher.
  • D may range from 3 to 9 dimensions or higher.
  • the high dimensional classifier projects the D dimensions for each iteration (different combination of features for each dimension) onto 1 dimension. Then, a separation metric is noted for that iteration (and set of features).
  • the high dimensional classifier selects the feature set that produced the highest separation metric. Along with the feature set, the projection and other information is stored for future inference. This is the output shown in the right-most block of Error! Reference source not found..
  • Error! Reference source not found further illustrates the high dimensional classifier referred to above.
  • N negative subjects in the training set and P positive subjects There are K features extracted and there are D dimensions and K is greater than D.
  • K features extracted For the negative set, a N by K matrix (N rows by K columns) of features is formed.
  • a P by K matrix (P rows by K columns) of features is formed.
  • the shaded columns in Error! Reference source not found represent a set of D features selected for a particular iteration.
  • PCA Principal Component Algorithm
  • the raw features may be pre-processed as shown in Error! Reference source not found.
  • the pre-processing is described in Error! Reference source not found..
  • Each column is concatenated into a single array for the purpose of normalization and standardization. The average of each combined column is subtracted from the column. Next, each of the combined columns are divided by their greatest quantile (for example, the value of the column at its 90 th percentile point). These quantiles (for each combined column) are stored for later inference (the process where disease detection is performed on individuals). After the division, the averages of each of the combined columns are re-added to their respective original (non-combined) columns.
  • the training process maximizes the separation metric between disease positive and negative subjects.
  • the high dimensional projection along with locations of extreme bins in the HDH's and all sea li ng/shifti ng information are stored for the inference process in which the training is applied to a new subject to determine disease status.
  • the inference step is shown in Error! Reference source not found..
  • Temperature features from multiple Regions of Interest (ROI) on the face can be combined for the detection of or contribute to the classification of the state of human health including diseases such as but not limited to COVID-19, influenza, streptococcus, and respiratory syncytial virus.
  • temperature data associated with each ROI mask is tabulated and analyzed for target features of interest.
  • the array of temperature features including but not limited to means, variances, quartile, etc. are used for training as shown in Error! Reference source not found, and for inference as shown in Error! Reference source not found..
  • Temperature spatial gradients within individual ROIs on the face can be used for the detection of or contribute to the classification of the state of human health including such as but not limited to COVID-19, influenza, streptococcus, and respiratory syncytial virus.
  • the purpose of this process is to measure the spatial structure inside individual ROIs in the face.
  • the reasoning is that temperatures will be elevated differently inside a subject when they have COVID19, flu or other infection. For example, obtaining a temperature by scanning the forehead only gives partial information; fever can be caused by quite a number of causes.
  • ambient temperature can also affect raw surface temperature readings of the forehead and the cheeks.
  • these spatial gradients can still contain useful signs of various diseases.
  • Spectral features extracted from hyperspectral images can be used for the detection of or contribute to the classification of the state of human health including diseases such as but not limited to COVID-19, influenza, streptococcus, and respiratory syncytial virus.
  • the schematic of this procedure is shown in Error! Reference source not found, which shows a hyperspectral image of a mouth and the corresponding spectral intensity which can also be applied to the eyes or other regions of interest on the human body.
  • the ROI is selected using a facial image segmentation algorithm on the central projection image. This is used to extract the region on the dispersed projection containing the spectral data associated with that ROI. This spectral data is then processed and interrogated for features such as locations of peaks, slopes in predetermined regions, and other features.
  • This algorithm is not restricted to wavelengths in the visible range. While the visible range color is a good indicator for detection in certain cases, this approach can be applied to a wider range of wavelengths from the UV to long-range IR (LWIR). For blazed gratings used in the IR range, such a tool adds temperature analysis functionality from the corresponding spectra.
  • LWIR long-range IR
  • a variety of optical elements could be utilized for the measurement of spectral information including wavelength specific features contained in an image. Such elements may include color filters or dispersive optical components, for example; a computer generated phase grating, holographic grating, prism, grism, or any other dispersive optical elements which could be tuned and utilized for different applications such as the direct detection of bacteria or viruses.
  • Recorded images (visible, ultraviolet or infrared) of the face and or eyes can be used for the detection of or contribute to the classification of the state of human health including diseases such as but not limited to COVID-19, influenza, streptococcus, and respiratory syncytial virus by extracting features from blood vessels within the sclera and their immediate surroundings.
  • diseases such as but not limited to COVID-19, influenza, streptococcus, and respiratory syncytial virus by extracting features from blood vessels within the sclera and their immediate surroundings.
  • HDH High Dimensional Histograms
  • HDH also result in features that will compete with others in training. HDH by themselves are not novel.
  • the application HDH in the training and inference process is. Values in every bin in the HDH for a set of disease positive subjects is compared with a set of disease negative subjects. Bins that yield separation between these two classes are noted and participate in training and inference as features.
  • Also provided herein is a method to reduce the number of training samples needed to automatically detect diseases that is also transparent to human interpretation and enhancement.
  • ANN Artificial Neural Networks
  • ANNs can be trained to detect diseases using a wide variety of cameras and other sensors.
  • ANNs require large numbers of training subjects for any reasonable level of specificity and sensitivity.
  • ANNs cannot be investigated if diagnostics are required to determine why a detection failed to correctly classify a disease. This is because ANNs are a large collection of trained "weights" which are just numbers that cannot be mapped to a particular physical metric or pathology.
  • a subject In order to perform multi-sensor disease detection, a subject must stand, or sit, or be positioned in front of the system as shown in Error! Reference source not found..
  • the wide FOV 13 and narrow FOV 12 sensors send data in individual frames or in a continuous stream such as video to the computer 30 (Error! Reference source not found.). Representations of recorded image data is shown in Error! Reference source not found..
  • a heuristic face finding algorithm is used on the recorded data creating a bounding rectangle around the face in the top half of Error! Reference source not found..
  • the positioning system 20 may include visual and or audio cues for guiding human self-positioning such as a display, monitor or speaker.
  • the positioning system 20 may consist of a robotic positioning system, that includes multi-axis positioning control such as a mechanized elevator 23 and a gimbal 22 with pan, and tilt motion control as shown in Error! Reference source not found, are automatically adjusted until the subject is located within a specified region of interest in the wide FOV imaging sensor. This will localize the eye in the narrow field of view sensor close enough so that the positioning system can be finetuned until the bounding rectangle from a feature detection finder is located within a specified region of interest as shown in the bottom half of Error! Reference source not found..
  • a similar process can be used to collect data from the spectral sensor 14 looking at the face, mouth, eye or specified region of interest.
  • a non-contact system consisting of an array of integrated sensors from a common frame or platform which can jointly measure human vital signs in conjunction with conducting disease detection.
  • the sensor array includes at least one visible imaging sensor 12, at least one infrared imaging sensor 11, at least one spectral sensor 14, at least one light source 15, and at least one electronic and software controller.
  • This unique combination of sensors is able to measure a continuous time sequence of human vital signs including body temperature, respiratory pattern including inhalation and exhalation, respiratory rate, pulse rate, heart rate, blood oxygen ratio, blood flow and pressure for both systolic and diastolic measurements, height, and approximate weight.
  • the multiple sensor dataset using feature extraction used for disease detection as described in this embodiment can also be used to measure human vital signs.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Surgery (AREA)
  • Primary Health Care (AREA)
  • Physics & Mathematics (AREA)
  • Animal Behavior & Ethology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Biophysics (AREA)
  • Veterinary Medicine (AREA)
  • Molecular Biology (AREA)
  • Epidemiology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

L'invention concerne un système sans contact pour diagnostiquer rapidement des maladies spécifiques ainsi que pour mesurer des signes vitaux sans collecte et manipulation de matériaux biologiques dangereux. Le système fusionne des informations provenant de multiples capteurs indépendants et différents pour maximiser la précision de diagnostic de maladie ainsi que des signes vitaux. Le système peut être utilisé pour cribler des sujets entrant dans des lieux, ce qui permet d'éviter les maladies transmissibles, ou pour des visites cliniques régulières. L'invention concerne également l'application d'approches d'apprentissage automatique qui sont suffisamment transparentes pour l'évaluation humaine et nécessitent des échantillons d'apprentissage minimaux.
PCT/US2023/031824 2022-07-01 2023-09-01 Système de diagnostic sans contact Ceased WO2024006586A2 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US202263367529P 2022-07-01 2022-07-01
US63/367,529 2022-07-01

Publications (2)

Publication Number Publication Date
WO2024006586A2 true WO2024006586A2 (fr) 2024-01-04
WO2024006586A3 WO2024006586A3 (fr) 2024-02-22

Family

ID=89381364

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2023/031824 Ceased WO2024006586A2 (fr) 2022-07-01 2023-09-01 Système de diagnostic sans contact

Country Status (1)

Country Link
WO (1) WO2024006586A2 (fr)

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120321759A1 (en) * 2007-01-05 2012-12-20 Myskin, Inc. Characterization of food materials by optomagnetic fingerprinting
EP3294106B1 (fr) * 2015-05-12 2020-08-26 Zipline Health, Inc. Dispositifs permettant d'acquérir des informations de diagnostic médical et permettant la fourniture de services de santé à distance
WO2019046602A1 (fr) * 2017-08-30 2019-03-07 P Tech, Llc Intelligence artificielle et/ou réalité virtuelle pour l'optimisation/la personnalisation d'activité
US11615687B2 (en) * 2018-01-26 2023-03-28 University Of Cincinnati Automated identification and creation of personalized kinetic state models of an individual
US20210327562A1 (en) * 2020-04-20 2021-10-21 PredictMedix Inc. Artificial intelligence driven rapid testing system for infectious diseases

Also Published As

Publication number Publication date
WO2024006586A3 (fr) 2024-02-22

Similar Documents

Publication Publication Date Title
US12124960B2 (en) Learning apparatus and learning method
US9149179B2 (en) System and method for identifying eye conditions
KR20200004841A (ko) 셀피를 촬영하도록 사용자를 안내하기 위한 시스템 및 방법
CN113159227A (zh) 一种基于神经网络的痤疮图像识别方法、系统和装置
CN112837805B (zh) 基于深度学习的眼睑拓扑形态特征的提取方法
JP6521845B2 (ja) 心拍に連動する周期的変動の計測装置及び計測方法
CN110175501B (zh) 基于人脸识别的多人场景专注度识别方法
Yadav et al. Computer‐aided diagnosis of cataract severity using retinal fundus images and deep learning
Sakaguchi et al. Fundus image classification for diabetic retinopathy using disease severity grading
KR102162683B1 (ko) 비정형 피부질환 영상데이터를 활용한 판독보조장치
US20230162354A1 (en) Artificial intelligence-based hyperspectrally resolved detection of anomalous cells
CN117764957A (zh) 基于人工神经网络的青光眼图像特征提取的训练系统
Caya et al. Development of pupil diameter determination using tiny-yolo algorithm
CN117373110B (zh) 可见光-热红外成像的婴幼儿行为识别方法、装置及设备
CN117530654A (zh) 一种实时的双眼瞳孔检查系统及检测方法
Yudhana et al. Glucose content analysis using image processing and machine learning techniques
Leopold et al. Segmentation and feature extraction of retinal vascular morphology
WO2024006586A2 (fr) Système de diagnostic sans contact
Kumar et al. Improved Blood Vessels Segmentation of Retinal Image of Infants.
Odstrcilik et al. Analysis of retinal nerve fiber layer via Markov random fields in color fundus images
Srinivasan et al. Complex texture features for glaucomatous image classification system using fundus images
Yadahalli et al. Detection glaucoma using deep learning algorithms
CN117731244A (zh) 一种基于红外热成像的脊柱侧弯风险预警系统
Ahmed et al. Automated glaucoma detection based on lbp representation and glrlm feature extraction method
Kowshik et al. Precise and Prompt Identification of Jaundice in Infants using AI

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: 23832435

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 23832435

Country of ref document: EP

Kind code of ref document: A2