WO2011087807A2 - Système et procédé de dépistage de mélanome à distance - Google Patents

Système et procédé de dépistage de mélanome à distance Download PDF

Info

Publication number
WO2011087807A2
WO2011087807A2 PCT/US2010/061667 US2010061667W WO2011087807A2 WO 2011087807 A2 WO2011087807 A2 WO 2011087807A2 US 2010061667 W US2010061667 W US 2010061667W WO 2011087807 A2 WO2011087807 A2 WO 2011087807A2
Authority
WO
WIPO (PCT)
Prior art keywords
remote user
image
server
user
data
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/US2010/061667
Other languages
English (en)
Other versions
WO2011087807A3 (fr
Inventor
Isabelle Guyon
Stephen D. Barnhill
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.)
Health Discovery Corp
Original Assignee
Health Discovery Corp
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 Health Discovery Corp filed Critical Health Discovery Corp
Publication of WO2011087807A2 publication Critical patent/WO2011087807A2/fr
Publication of WO2011087807A3 publication Critical patent/WO2011087807A3/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • 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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/41—Detecting, measuring or recording for evaluating the immune or lymphatic systems
    • A61B5/414—Evaluating particular organs or parts of the immune or lymphatic systems
    • A61B5/415—Evaluating particular organs or parts of the immune or lymphatic systems the glands, e.g. tonsils, adenoids or thymus
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/41—Detecting, measuring or recording for evaluating the immune or lymphatic systems
    • A61B5/414—Evaluating particular organs or parts of the immune or lymphatic systems
    • A61B5/418—Evaluating particular organs or parts of the immune or lymphatic systems lymph vessels, ducts or nodes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/44—Detecting, measuring or recording for evaluating the integumentary system, e.g. skin, hair or nails
    • A61B5/441—Skin evaluation, e.g. for skin disorder diagnosis
    • A61B5/444—Evaluating skin marks, e.g. mole, nevi, tumour, scar
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6887—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient mounted on external non-worn devices, e.g. non-medical devices
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6887—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient mounted on external non-worn devices, e.g. non-medical devices
    • A61B5/6898—Portable consumer electronic devices, e.g. music players, telephones, tablet computers
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • 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
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/40—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
    • 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
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/67—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for remote operation
    • 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
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • 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
    • A61B5/0082—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence adapted for particular medical purposes
    • A61B5/0091—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence adapted for particular medical purposes for mammography

Definitions

  • the present invention relates to a system and method for computer image analysis for screening for skin cancer using photographic images transmitted over a distributed network from a remote location, and more particularly to a system and method for photographing and downloading images of potential melanoma lesions for processing using learning machines to provide a preliminary risk assessment of melanoma.
  • melanoma Malignant melanoma is currently one of the leading cancers among many light- skinned populations around the world. Changes in recreational behavior together with the increase in ultraviolet radiation due to thinning or lost of the earth's ozone layer have caused a dramatic increase in the number of melanomas diagnosed. The rise in incidence was first noticed in the United States in 1930, where one person out of 100,000 per year suffered from skin cancer. This rate increased in the mid-eighties to six per 100,000 and to 13 per 100,000 in 1991. In fact, melanoma is currently one of the most common cancers in young adults. Each year, more than 50,000 people in the U.S. learn that they have melanoma. According to the World Health Organization website, 132,000 new cases of melanoma skin cancer occur globally each year.
  • Melanoma starts in the pigment-producing skin cells (melanocytes).
  • the first sign of melanoma is often a change in the size, shape, or color of an existing mole or the appearance of a new mole. Since the vast majority of primary melanomas are visible on the skin, there is a good chance of detecting the disease in its early stages. If not detected at treated at an early stage, these cells become abnormal, grow uncontrollably, and aggressively invade surrounding tissues.
  • Melanoma can spread quickly and produce large malignant tumors in the brain, lung, liver, or other organs, with depth of penetration being predictive of prognosis: Epidermis only: Clark level I. Upper dermis: Clark levels II and II. Lower dermis: Clark level IV. Fatty layers: Clark level V.
  • Non-melanomas are the most common cancers of the skin. Because they rarely spread elsewhere in the body, they are less worrisome than melanomas. Melanoma is much less common than basal cell and squamous cell skin cancers, but it is far more serious. Because it begins in the melanocytes, most of these cells keep on making melanin thus melanoma tumors are often brown or black (but not always).
  • the gold standard for accurate diagnosis remains histological examination of biopsies.
  • the type of biopsy depends on the size of the skin growth and its location on the body. Several types of biopsy can be done when melanoma is suspected. The first is an excisional biopsy, which cuts away the entire growth with a margin of normal surrounding skin. A second type is an incisional biopsy, or core biopsy, removing only a sample of the growth. A punch biopsy removes a small, cylindrical shaped sample of skin.
  • a fourth type is a saucerization biopsy, which removes the entire lesion by cutting under the lesion in a "scoop like" manner.
  • a fifth type is a fine -needle aspiration biopsy done with a very thin needle, which removes a very small sample of tissue (usually not done on moles but on other deeper tissue, such as nearby lymph nodes). Prognosis is assessed by the TNM system (T stands for tumor thickness and how far it has spread; N stands for lymph nodes, and whether the tumor has spread to the nodes; and M stands of metastasis, and whether the tumor has spread to distant organs).
  • melanoma may also be diagnosed, to some extent, from the appearance of the skin surface.
  • Four main features of the appearance are used: asymmetry, uneven edges, multiple shades, and size.
  • These characteristics known as the "ABCD" characteristics, provide a subjective means for physicians and patients to identify pigmented skin lesions that could be melanoma.
  • the four parameters represented by the ABCD characteristics are lesion asymmetry (A), border irregularity (B), color variegation (C) and lesion diameter (D).
  • A lesion asymmetry
  • B border irregularity
  • C color variegation
  • D lesion diameter
  • experienced dermatologists can identify a melanoma with around 75% accuracy (Serruys, 1999).
  • the MelaFind ® scanner is a large hand-held scanner housing a multi-spectral light source and a sensor that is placed directly in contact with the lesion.
  • the MelaFind scanner is designed for use by medical professionals and is not intended for general consumer use, which means that the patient must have already suspected a problem and consulted a physician before such a scanner would be available for use on the patient.
  • Reported efforts to develop methods for machine -based diagnosis of melanoma using digital images include a number of pre-processing steps, such as standardizing illumination, shading correction, noise filtering for color quality and use of polarizing filters.
  • the image resolution varies from study to study, but typically is not lower than 256x256 pixel images, with 0.01 cm/pixel and 24 bit per pixel color depth.
  • Accessibility of machine-based diagnosis can be extended by using everyday digital images, such as images taken using the built-in camera of a smart phone or a simple digital camera. Such an approach would make melanoma screening more accessible to individuals who are concerned about the health of their skin but have not yet been able to consult a physician. However, the quality of such images tends to be fairly low.
  • Optimized extraction and reconstruction of data within an image can be problematic where sources of noise and other factors can negatively impact the ability to efficiently extract data from the image, thus impairing the effectiveness of the imaging method for its intended use.
  • Examples of areas in which image analysis can be problematic include astronomical observation and planetary exploration, where sources can be faint and atmospheric interference introduce noise and distortion, military and security surveillance, where light can be low and rapid movement of targets result in low contrast and blur, and medical imaging, which often suffers from low contrast, blur and distortion due to source and instrument limitations.
  • Adding to the difficulty of image analysis is the large volume of data contained within a digitized image, since the value of any given data point often cannot be established until the entire image is processed.
  • Machine-learning approaches for image analysis have been widely explored for recognizing patterns which, in turn, allow extraction of significant features within an image from a background of irrelevant detail.
  • Learning machines comprise algorithms that may be trained to generalize using data with known outcomes. Trained learning machine algorithms may then be applied to predict the outcome in cases of unknown outcome.
  • Machine-learning approaches which include neural networks, hidden Markov models, belief networks and support vector machines, are ideally suited for domains characterized by the existence of large amounts of data, noisy patterns and the absence of general theories.
  • chromosome images for identifying genetic abnormalities, and tumor detection in ultrasound images, among others.
  • the majority of learning machines that have been applied to image analysis are neural networks trained using back-propagation, a gradient-based method in which errors in classification of training data are propagated backwards through the network to adjust the bias weights of the network elements until the mean squared error is minimized.
  • a significant drawback of back-propagation neural networks is that the empirical risk function may have many local minimums, a case that can easily obscure the optimal solution from discovery. Standard optimization procedures employed by back-propagation neural networks may converge to a minimum, but the neural network method cannot guarantee that even a localized minimum is attained, much less the desired global minimum. The quality of the solution obtained from a neural network depends on many factors.
  • the skill of the practitioner implementing the neural network determines the ultimate benefit, but even factors as seemingly benign as the random selection of initial weights can lead to poor results.
  • the convergence of the gradient-based method used in neural network learning is inherently slow.
  • the sigmoid function has a scaling factor, which affects the quality of approximation.
  • the largest limiting factor of neural networks as related to knowledge discovery is the "curse of dimensionality" associated with the disproportionate growth in required computational time and power for each additional feature or dimension in the training data.
  • a support vector machine maps input vectors into high dimensional feature space through a non- linear mapping function, chosen a priori.
  • an optimal separating hyperplane is constructed.
  • the optimal hyperplane is then used to perform operations such as class separations, regression fit, or density estimation.
  • SVMs are well-recognized as having the advantage in solving classification problems of high dimension and small size dataset.
  • U.S. Patent Nos. 6,157,921, 6,714,925, and 7,797,257 which are incorporated herein by reference, describe a system and method for providing SVM analysis services for processing of data transmitted from a remote source over the Internet to a processor that executes trained SVMs.
  • the processor receives the data from the remote source along with account information that provides for a financial transaction to secure payment for the analysis services.
  • the analysis results are transmitted to the remote requestor over the Internet and a transaction is initiated, for example with a financial institution, to secure payment for the data analysis services from the designated account.
  • the present invention relates to systems and methods for enhancing knowledge discovered from data using a learning machine in general, and a support vector machine in particular.
  • the present invention comprises methods of using a learning machine for diagnosing and prognosing changes in biological systems such as diseases. Further, once the knowledge discovered from the data is determined, the specific relationships discovered are used to diagnose and prognose diseases, and methods of detecting and treating such diseases are applied to the biological system.
  • One embodiment of the present invention comprises preprocessing a training data set in order to allow the most advantageous application of the learning machine.
  • Each training data point comprises a vector having one or more coordinates.
  • Preprocessing the training data set may comprise identifying missing or erroneous data points and taking appropriate steps to correct the flawed data or as appropriate remove the observation or the entire field from the scope of the problem.
  • Pre-processing the training data set may also comprise adding dimensionality to each training data point by adding one or more new coordinates to the vector.
  • the new coordinates added to the vector may be derived by applying a transformation to one or more of the original coordinates.
  • the transformation may be based on expert knowledge, or may be computationally derived. In a situation where the training data set comprises a continuous variable, the transformation may comprise optimally categorizing the continuous variable of the training data set.
  • the support vector machine is trained using the pre- processed training data set.
  • the additional representations of the training data provided by the preprocessing may enhance the learning machine's ability to discover knowledge therefrom.
  • the greater the dimensionality of the training set the higher the quality of the generalizations that may be derived therefrom.
  • the training output may be post-processed by optimally categorizing the training output to derive categorizations from the continuous variable.
  • a test data set is pre-processed in the same manner as was the training data set. Then, the trained learning machine is tested using the pre-processed test data set.
  • a test output of the trained learning machine may be post-processed to determine if the test output is an optimal solution.
  • Post-processing the test output may comprise interpreting the test output into a format that may be compared with the test data set. Alternative postprocessing steps may enhance the human interpretability or suitability for additional processing of the output data.
  • a method for the selection of at least one kernel prior to training the support vector machine is also disclosed.
  • the selection of a kernel may be based on prior knowledge of the specific problem being addressed or analysis of the properties of any available data to be used with the learning machine and is typically dependant on the nature of the knowledge to be discovered from the data.
  • an iterative process comparing postprocessed training outputs or test outputs can be applied to make a determination as to which configuration provides the optimal solution. If the test output is not the optimal solution, the selection of the kernel may be adjusted and the support vector machine may be retrained and retested.
  • a live data set may be collected and pre-processed in the same manner as was the training data set.
  • the pre-processed live data set is input into the learning machine for processing.
  • the live output of the learning machine may then be post- processed by interpreting the live output into a computationally derived alphanumeric classifier or other form suitable to further utilization of the SVM-derived answer.
  • a system for enhancing knowledge discovered from data using a support vector machine.
  • the exemplary system comprises a storage device for storing a training data set and a test data set, and a processor for executing a support vector machine.
  • the processor is also operable for collecting the training data set from the database, pre-processing the training data set to enhance each of a plurality of training data points, training the support vector machine using the pre-processed training data set, collecting the test data set from the database, pre-processing the test data set in the same manner as was the training data set, testing the trained support vector machine using the pre-processed test data set, and in response to receiving the test output of the trained support vector machine, post-processing the test output to determine if the test output is an optimal solution.
  • the exemplary system may also comprise a communications device for receiving a live data set from a remote source.
  • the processor may be operable to store the live data set in the storage device prior pre-processing.
  • the exemplary system may also comprise a display device for displaying the post-processed results.
  • the processor of the exemplary system may further be operable for performing each additional function described above.
  • the communications device may be further operable to send a computationally derived alphanumeric classifier or other raw or post-processed output data to a remote source.
  • the processor may be operable to communicate with a financial institution or other account provider for the purpose of securing payment for analysis services through an account identified by an account identifier provided by the service requester.
  • a system and method are provided for enhancing knowledge discovery from data using multiple learning machines in general and multiple support vector machines in particular.
  • Multiple support vector machines each comprising distinct kernels, are trained with the pre-processed training data and are tested with test data that is pre- processed in the same manner.
  • the test outputs from multiple support vector machines may be compared in order to determine which of the test outputs, if any, represents an optimal solution.
  • Selection of one or more kernels may be adjusted and one or more support vector machines may be retrained and retested.
  • live data is pre-processed and input into the support vector machine comprising the kernel(s) that produced the optimal solution.
  • the live output from the learning machine may then be post-processed into a computationally derived alphanumeric classifier for interpretation by a human or computer automated process.
  • a system and method are provided for enhancing knowledge discovery from data using a learning machine that is accessible via a distributed network environment, e.g., the Internet.
  • a customer may transmit training data, test data and/or live data to a central server from a remote source over the network.
  • the customer may also transmit to the server identification information such as a user name, a password, geographical location, an account identifier, or the financial account identifier of a third party's account, charges to which are initiated by information entered by the user.
  • the account identifier is associated with an Internet-enabled smart phone, such as a mobile phone number or wireless service account number, so that billing for analysis services can be charged to the user's mobile phone account.
  • the training data, test data and/or live data may be stored in a storage device at the central server.
  • the learning machine is trained and tested prior to receiving and processing of live data that is transmitted by the remote user.
  • Training data may be pre-processed in order to add meaning thereto.
  • Preprocessing data may involve transforming the data points and/or expanding the data points. By adding meaning to the data, the learning machine is provided with a greater amount of information for processing.
  • the learning machine which may be a support vector machine, a random forest classifier, a Gaussian classifier or other classifier, or an ensemble classifier, is trained with the pre-processed training data and is tested with test data that is pre-processed in the same manner.
  • the test output from the learning machine is post-processed in order to determine if the knowledge discovered from the test data is desirable. In other words, the output is evaluated to determine if the correct classification has been made by the learning machine.
  • Post-processing involves interpreting the test output into a format that may be compared with the test data. Once the learning machine has been satisfactorily trained, live data is pre-processed and input into the trained and tested learning machine. The live output from the learning machine may then be post-processed into a computationally derived alphanumerical classifier or converted into a graphical display for easy interpretation by a human.
  • a system for analyzing image data received from a remote user for evaluating an image for screening for a disease or condition, the system comprising: a server in communication with a distributed network for receiving a digital image data set from the remote user, the remote user also in communication with the distributed network; a processor for executing a learning machine, wherein the learning machine is trained using image data sets having known outcomes for skin cancer, the processor further operable for: receiving the digital image data set from the remote user; pre-processing the digital image data set to extract from the image; inputting the extracted features into the trained learning machine to produce an output comprising a recognized pattern within the digital image data set; post- processing the output to generate a score corresponding to the recognized pattern associated with the disease or condition; and transmitting the score to the server;
  • server is further operable for transmitting the score to the remote user across the distributed network.
  • a system and method for analyzing data comprising a digital image taken by an individual at a remote location who wishes to obtain a preliminary screening for skin cancer such as melanoma.
  • a system for analyzing image data received from a remote user for screening for skin cancer, the system comprising: a server in communication with a distributed network for receiving a digital image data set from the remote user, the remote user also in communication with the distributed network; a processor for executing a learning machine, wherein the learning machine is trained using image data sets having known outcomes for skin cancer, the processor further operable for: receiving the digital image data set from the remote user; preprocessing the digital image data set to extract features including contour, dimension and color features; inputting the extracted features into the trained learning machine to produce an output comprising a recognized pattern within the digital image data set; post-processing the output to generate a skin cancer risk score corresponding to the recognized pattern; and transmitting the skin cancer risk score to the server; wherein the server is further operable for transmitting the alphanumerical skin cancer risk score to the remote user across the distributed network.
  • the learning machine located at a central server accessible via the Internet is trained and tested for classifying melanoma images using image data obtained and downloaded by the service provider, while the live data is provided by a remote user who is interested in receiving a preliminary screening for melanoma using an automated, computer-based analysis.
  • the sources of the image data for training and testing of the learning machine may include medical literature and image databases on the Internet, diagnostic laboratories and research institutions.
  • the image data will have known classifications based on expert evaluations, i.e., by a pathologist or dermatologist, or a combination of a visual evaluation by a dermatologist and confirmation by a pathologist using histological methods.
  • a method for diagnosing melanoma from digital images taken with a multi-media enabled smart phone or a digital camera and transmitted to a remote central server, e.g., by email or by download to a dedicated website associated with the central server.
  • the user may be requested to enter additional information that can be combined with the image data in the classification process.
  • the user-entered information may include size, shape, color, itching, bleeding, and/or changes over time.
  • the user may also be requested to download additional photographs of other, less worrisome markings (moles) on their skin for comparison.
  • the image is pre-processed to enhance the image quality.
  • Pre-processing may involve actions by the user in response to instructions provided by the system that are intended to optimize the clarity and accuracy of the image. Following user-pre-processing and download, pre-processing may include one or more of segmentation, extraction of contours of inner structure, extraction of geometrical features, and extraction of color features. The features that are evaluated are based on the ABCD scale that is commonly used by dermatologists to diagnose melanoma: asymmetry ("A"), border ("B”), color (“C”), and diameter (“D”).
  • A asymmetry
  • B border
  • C color
  • D diameter
  • Appropriate learning machines for such a task include support vector machines, neural networks, random forests, Bayesian classifiers, other statistically-based classifiers, or other classifiers, or combinations thereof.
  • a kernel based machine such as a support vector machine is used.
  • an ensemble of classifiers is used, with each classifier being trained on a different feature set.
  • each classifier is postprocessed to obtain a mapping of outputs to probabilities.
  • an ensemble of classifiers is obtained by voting of different base classifiers, with each base classifier being given an equal weight.
  • a second level, overall or stacking classifier that has been trained to generate a single "diagnosis" based on inputs consisting of the outputs of the different feature classifiers, i.e., ABCD, receives the result from each classifier and generates a score.
  • the resulting vote (score) of the ensemble or second level classifier is post- processed to obtain a mapping of the output to probabilities.
  • the output is converted into an alphanumeric and/or graphical display that may be stored in a memory medium and/or transmitted to the remote user to provide an overall probability, i.e., a confidence level, that the lesion in the image is melanoma.
  • explanatory language may be included within the transmitted graphical display.
  • referral suggestions may be provided based upon the user's geographical location, which may be obtained from a smart phone with location services, e.g., GPS capability, if available and accessible to the central server, or by requesting that the user input information such as the zip code, area code or city/state/country in which they are located.
  • location services e.g., GPS capability
  • the central server can access and search a database of qualified physicians or medical facilities within the same city or geographical area that can provide further evaluation and/or treatment of the user's possible condition. If the system's analysis of the transmitted photograph of the area of concern indicates an elevated risk level, the system will identify one or more physicians within the sender's area and will transmit contact information for the physician at the same time the sender's risk level is sent.
  • the sender will be charged a nominal fee for the analysis of the transmitted image by requesting credit or debit card information or providing some other on-line payment mechanism such as PayPal ® or similar Internet-based transaction services.
  • the billing may be enabled via a contract between the analysis service provider and the wireless service provider, e.g., as a data charge that can be included on the user's wireless billing statement.
  • a typical fee may be on the order of $2 to $20.
  • the rationale for charging such a fee would be that the persons using the service would genuinely be interested in obtaining the analysis, as opposed to people who are simply taking advantage of a free service or "playing", without having a genuine need for the screening services.
  • the database can instead include listings of physicians who will be charged a fee for each referral. When a particular physician is recommended to the remote user, that physician would be charged a fee.
  • the service could be fully paid for by a diagnostic laboratory and the database would include a listing of physicians who contract with that diagnostic laboratory for skin cancer diagnostics such that the diagnostic laboratory would benefit by ultimately receiving the request for pathology services and billing the patient or their insurance provider.
  • FIG. 1 is a flow diagram of an exemplary method according to the invention.
  • FIG. 2 is a functional block diagram of a network based system for providing melanoma screening.
  • FIG. 3 is a block diagram of the basic architecture for a system according to the invention.
  • FIG. 4 is a functional block diagram of an exemplary operating environment for an embodiment of the present invention.
  • FIG. 5 is a functional block diagram of an exemplary network operating environment for implementation of the present invention.
  • FIG. 6 is a block diagram showing additional details of an exemplary implementation of the image preprocessing and feature selection components of the system.
  • FIG. 7 is a digital image of an exemplary suspected melanoma showing the outer and inner contours extracted by the segmentation algorithm.
  • FIGs. 8a and 8b are plots used for measurement of the geometrical features of the suspected melanoma, where FIG. 8a shows the center of gravity of the image and the average radius for determining coefficient of variance of the radius, and FIG. 8b shows the distances from the center of gravity used to determine radius aspect.
  • FIGs. 9a and 9b show the outer and inner contours, respectively, for comparison of color within and outside the contour.
  • FIG. 10 is a gray scale feature matrix of an image dataset based on 20 features in which the upper half corresponds to controls and the lower half corresponds to malignant melanoma.
  • FIG. 11 is a receiver operating characteristic (ROC) curve for 10 x 10 cross- validation of the image classifier.
  • FIG. 12 is a sample report displayed to the user following classifier analysis of the image of FIG. 7.
  • FIG. 13 is a functional block diagram of a hierarchical or ensemble classifier system.
  • FIG. 14 is a functional block diagram of a multilevel classifier system for remote melanoma screening according to the present invention.
  • FIG. 15 is a block diagram of an exemplary multi-module construction for a remote melanoma screening system and method.
  • FIG. 16 is a plot of a sigmoid function for classification of mole or freckle diameter according to one embodiment of the invention.
  • FIGs. 17a-17d are images of a smart phone with sample displays produced according to one embodiment of the invention.
  • FIG. 18 is an image of a smart phone with an alternative survey display.
  • FIGs. 19a- 19c are exemplary geometric constructs for feature extraction for use in screening out garbage images.
  • FIG. 20 is a flow diagram of an embodiment of the risk analysis system of the present invention.
  • the present invention provides methods, systems and devices for discovering knowledge from data using learning machines.
  • the present invention is directed to methods, systems and devices for knowledge discovery from data using learning machines that are provided information regarding changes in biological systems. More particularly, the present invention comprises methods of use of such knowledge for diagnosing and prognosing changes in biological systems such as diseases. Additionally, the present invention comprises methods, compositions and devices for applying such knowledge to the testing and treating of individuals with changes in their individual biological systems.
  • biological data means any data derived from measuring biological conditions of human, animals or other biological organisms including microorganisms, viruses, plants and other living organisms. The measurements may be made by any tests, assays or observations that are known to physicians, scientists, diagnosticians, or the like. Biological data may include, but is not limited to, clinical tests and observations, including medical images, physical and chemical measurements, genomic determinations, proteomic determinations, drug levels, hormonal and immunological tests, neurochemical or neurophysical measurements, mineral and vitamin level determinations, genetic and familial histories, and other determinations that may give insight into the state of the individual or individuals that are undergoing testing.
  • data is used interchangeably with “biological data”.
  • learning machines comprise algorithms that may be trained to generalize using data with known outcomes. Trained learning machine algorithms may then be applied to cases of unknown outcome for prediction. For example, a learning machine may be trained to recognize patterns in data, estimate regression in data or estimate probability density within data. Learning machines may be trained to solve a wide variety of problems as known to those of ordinary skill in the art. A trained learning machine may optionally be tested using test data to ensure that its output is validated within an acceptable margin of error. Once a learning machine is trained and tested, live data may be input therein. The live output of a learning machine comprises knowledge discovered from all of the training data as applied to the live data.
  • the present invention provides a system and method for diagnosing melanoma from images taken with a smart phone, such as the iPhone ® (Apple, Inc.), RIM Blackberry ® , Windows ® Mobile, Google ® Android, and similar mobile phones with PC-like functionality and cameras, and transmitted to a central server, e.g., by email or download to a website.
  • the inventive method includes an approach in which the quality of the smart phone images could be improved by helping the user to take better pictures and crop the area of interest.
  • the user may be asked to provide information obtained from self examination, such as changes in size, shape, color, itching, or pictures of other less worrying moles.
  • the result can be provided in an educational form to assist the patient in understanding the diagnosis and decide whether to consult a doctor.
  • FIG. 1 provides a high level flow diagram showing the modular components of the screening process using a smart phone with multi-media capability and Internet access according the present invention.
  • the person wishing to obtain a melanoma screening provides requested information by accessing a website linked to an image analysis server.
  • the requested information may include a series of questions that will better enable the image analysis and classification process to evaluate the requester's risk of melanoma.
  • the questions may include dimensional information, how long ago the suspect lesion was noticed, any changes, etc. If the requester is using a smart phone, he or she will be instructed to take a photograph of the suspected melanoma lesion using the built-in camera of the smart phone.
  • Instructions may be provided to the requester to optimize the image by cropping or improving the lighting conditions.
  • a digital camera may be used and the jpeg or other file format downloaded and saved on the smart phone or a personal computer or laptop.
  • the requester transmits the digital photograph to the central server by e-mail, or alternatively by downloading the image to a website associated with the server.
  • An image analysis server that is programmed to execute pre-processing algorithms for extracting relevant features and a classifier that has been trained to distinguish between melanoma and other conditions is used to process the input image (step 108) to assign the suspected cancer to one of a small number of risk categories ranging from low to high, or to provide a probability which is considered a "risk score".
  • a report in the form of the risk score is transmitted to the smart phone or to the user's e- mail address, or is made available on-line with password protection.
  • classifiers that may be appropriate for use in generating a risk score based on analysis of different features of interest within an image can be found in U.S. Patents No.
  • the risk score may be accompanied by a table explaining the scoring levels, and additional information may be provided about melanoma.
  • the central server will search a database to provide the requester with referrals to a physician who specializes in treatment of skin cancers.
  • the present invention is not intended to provide a definitive diagnosis or not of melanoma; it is merely intended to serve as a simple, low-cost preliminary screening tool to allow a person to obtain an advisory indication of whether he or she should seek a formal medical evaluation.
  • the requester in this example, a smart phone user, uses the camera function of the smart phone 202 to take a photograph of the suspected melanoma site 204 and selects the phone's e-mail function to transmit the photograph over the Internet 206 to the system's central server 210 along with identifying information, such as a user name or number.
  • identifying information such as a user name or number.
  • the user will be requested to respond to a number of survey questions that will assist in the classification.
  • the user's identity will not be utilized, but an account number or password may be assigned.
  • payment information e g., a credit card number, telephone number or other account information may be collected.
  • the central server 210 receives the photograph, survey responses, and, if appropriate, the sender's identification and account information for payment.
  • the central server may also request or automatically obtain the sender's geographic location. Location information may be obtained through the smart phone's GPS function, which may be obtained automatically if the sender has allowed unlimited access to their GPS information, or the sender may have granted permission to access their GPS function in response to a request sent by the central server.
  • the sender's area code can be obtained from the caller ID, or the server can request that the sender enter their area or zip code.
  • the central server 210 may respond to the user with instructions to take the photograph at a shorter distance from the suspected lesion, to utilize additional lighting, e.g., a flash or to move closer to a light source, or to modify the original image by cropping it. These operations generally fall within step 302 (shown in FIG. 4) of the pre-processing block 214 even though they are not actually performed by the server.
  • the pre-processing block 214 may include an algorithm for screening spurious or "garbage" images that are not moles and could be either obviously bad data or a joke by a user who is playing with the screening service.
  • FIG. 3 provides a high-level block diagram of an embodiment of the modular system architecture for implementing the melanoma screening platform according to the present invention.
  • the images may be captured either with a smart phone camera 202 or with a camera 404 and downloaded to a personal computer 406.
  • the camera 404 is preferably digital, but if not, the film images can be scanned using a color scanner and downloaded to the computer 406.
  • an upload protocol will be implemented.
  • the first protocol will be for a smart phone, e.g., an iPhone ® , with dedicated image capture software and data transmission to the image analysis server 410.
  • the second protocol will be for transmission by e-mail of an image captured using a standard mobile phone, digital camera or scanner.
  • the third protocol will be an upload program on a website for images captured by a mobile phone, digital camera or scanner.
  • Answers to a short survey completed by the user may also be transmitted to the image analysis server 410 with the image and its meta data.
  • the server may include programming to check continuously for incoming images.
  • the images may be in a standard image format, such as jpeg, tiff or other format, while the answers to the survey may be an ASCII or XML format file.
  • a user may be pre-registered, which will allow them to upload data via either a smart phone application or the website.
  • the upload application whether on the smart phone or on the website, will identify the user (patient), store the two files (image and text) using a specified file nomenclature.
  • the file names can be patientID_dattime.jpg and patientID_datetime.txt.
  • the user will be responsible for providing images of sufficient resolution, with good lighting and focus, following guidelines provided on the website. Guidelines may include suggested distance between the camera and the suspected melanoma and lighting conditions.
  • the application software can transmit messages to the smart phone with guidance to the user such as indicating whether the lighting is sufficient, indicating whether the distance to the skin is appropriate to obtain proper focus, and helping the user crop the image.
  • Suggestions may include finding an object to support the camera or smart phone to assist in holding the camera steady and parallel to the lesion. Ideally, the camera should be held 10-15 cm above the lesion.
  • Instructions may include zooming in on the lesion using the touch screen slider to enlarge it to the maximum that fits in the screen, leaving a small border. A box or rectangle may be displayed on the screen to further assist the user.
  • the smart phone application may also transmit the location recorded with the
  • GPS in the photos assuming the "location services” function is turned on. Instead, if going through “Settings” - “General", obtaining the GPS information can be incorporated into the coding of the application.
  • Some smart phones with built-in GPS receivers are able to encode location information directly in the EXIF (exchangable image file format) meta data associated with the transmitted image. All EXIF meta data will be transmitted to the central server, which can provide autofocus information, which may be used to determine image scale. In some smart phones, the EXIF data is automatically stripped away when the photos are e-mailed, so the application must include provisions to allow the EXIF data to be transmitted.
  • HostComputer 'Mac OS X 10.5.8'
  • SensingMethod 'One-chip color area sensor' info.GPSInfo
  • GPSAltitudeRef 0
  • GPS Altitude 304
  • the upload program will also interact with the database server 430 to identify the user and notify the database server 430 of the upload of image(s) and survey responses.
  • the patient and administrator database on database server 430 may be a mySQL database for holding the user database.
  • the database server will be accessible to two types of users: patients and administrators.
  • the user table may include basic demographic information: Name, Address, Email (which can serve as the user's login), password. Users may be registered with a unique ID which can be automatically generated.
  • Patient user privileges on the database server 430 can include uploading images and accessing their own "My Lab" space on the website. Administrator user privileges may include patient user privileges plus access to the administrative backend to allow editing of website contents and management of the patient user database.
  • the database server 430 must have scalable capacity. It need not physically be a separate server structure— it may be the same as the web server 420.
  • the transmitted images may be stored directly on the image analysis server 410 as flat files. As soon as the images and surveys are uploaded, they can be processed by image analysis server 410.
  • the image analysis software may be implemented in Matlab (Mathworks, Natick, MA), however, a faster platform may be preferred for performance reasons.
  • an exemplary image analysis server may include the following features: a Microsoft ® Windows ® 2003 server with IIS, 2-4 Gbytes of RAM, 500 Gbytes of disk space, upgradable, to store images.
  • the server configuration requirements may include: allow RSH, configure as web server, configure as ftp server, install mySQL and PHP, install Perl, install Matlab ® with the Statistics, Image Processing and Optimization toolboxes (also from Mathworks).
  • An alternative server would be a Linux Redhat Enterprise 5 Server with the Apache HTTP server software.
  • the image analysis server 410 preferably includes storage for all images and surveys and interacts with the database server 430 and the web server 420, including notifying the database server 430 when the image analysis is completed.
  • the web server 420 will serve static pages with various types of information, for example, statistics on melanoma, as well as dynamic pages, which will include the analysis results and physician referral pages and/or links based on the user's geographical location information.
  • the web server 420 may also include advertisements for physicians or for skin care products such as sun screen and other protective skin treatments.
  • FIG. 4 and the following discussion are intended to provide a brief and general description of a suitable computing environment for implementing one aspect of the present invention.
  • the computer 1000 includes a central processing unit 1022, a system memory 1020, and an Input/Output ("I/O") bus 1026.
  • a system bus 1021 couples the central processing unit 1022 to the system memory 1020.
  • a bus controller 1023 controls the flow of data on the I/O bus 1026 and between the central processing unit 1022 and a variety of internal and external I/O devices.
  • the I/O devices connected to the I/O bus 1026 may have direct access to the system memory 1020 using a Direct Memory Access (“DMA”) controller 1024.
  • DMA Direct Memory Access
  • the I/O devices are connected to the I/O bus 1026 via a set of device interfaces.
  • the device interfaces may include both hardware components and software
  • a hard disk drive 1030 and a floppy disk drive 1032 for reading or writing removable media 1050 may be connected to the I/O bus 1026 through disk drive controllers 1040.
  • An optical disk drive 1034 for reading or writing optical media 1052 may be connected to the I/O bus 1026 using a Small Computer System Interface ("SCSI") 1041.
  • SCSI Small Computer System Interface
  • an IDE (AT API) or EIDE interface may be associated with an optical drive such as a may be the case with a CD-ROM drive.
  • the drives and their associated computer-readable media provide nonvolatile storage for the computer 1000.
  • other types of computer-readable media may also be used, such as ZIP drives or removable media such as flash drives or the like.
  • a display device 1053 such as a monitor, is connected to the I/O bus 1026 via another interface, such as a video adapter 1042.
  • a parallel interface 1043 connects synchronous peripheral devices, such as a laser printer 1056, to the I/O bus 1026.
  • a serial interface 1044 connects communication devices to the I/O bus 1026.
  • a user may enter commands and information into the computer 1000 via the serial interface 1044 or by using an input device, such as a keyboard 1038, a mouse 1036 or a modem 1057.
  • Other peripheral devices may also be connected to the computer 1000, such as audio input/output devices or image capture devices.
  • a number of program modules may be stored on the drives and in the system memory 1020.
  • the system memory 1020 can include both Random Access Memory (“RAM”) and Read Only Memory (“ROM”).
  • the program modules control how the computer 1000 functions and interacts with the user, with I/O devices or with other computers.
  • Program modules include routines, operating systems 1065, application programs, data structures, and other software or firmware components.
  • the present invention may comprise one or more pre-processing program modules 1075 A, one or more post-processing program modules 1075B, and/or one or more optimal categorization program modules 1077 and one or more SVM program modules 1070 stored on the drives or in the system memory 1020 of the computer 1000.
  • pre-processing program modules 1075 A, post-processing program modules 1075B, together with the SVM program modules 1070 may comprise computer-executable instructions for pre-processing data and post-processing output from a learning machine and implementing the learning algorithm according to the exemplary methods described herein.
  • optimal categorization program modules 1077 may comprise computer-executable instructions for optimally categorizing a data set.
  • the computer 1000 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer 1060.
  • the remote computer 1060 may be a server, a router, a peer device or other common network node, and typically includes many or all of the elements described in connection with the computer 1000.
  • program modules and data may be stored on the remote computer 1060.
  • the logical connections depicted in FIG. 4 include a local area network ("LAN”) 1054 and a wide area network (“WAN”) 1055.
  • a network interface 1045 such as an Ethernet adapter card, can be used to connect the computer 1000 to the remote computer 1060.
  • the computer 1000 may use a telecommunications device, such as a modem 1057, to establish a connection.
  • a telecommunications device such as a modem 1057
  • the network connections shown are illustrative and other devices of establishing a communications link between the computers may be used.
  • FIG. 5 is a functional block diagram illustrating an exemplary network operating environment for implementation of the present invention.
  • a remote user 1202 or other entity may transmit data via a distributed computer network, such as the Internet 1204, to a service provider 1212, e.g., a website host, who provides analysis services as described above.
  • a service provider 1212 e.g., a website host
  • the customer 1202 may transmit data from any type of computer, laboratory instrument or multi-media device, including a smart phone, with network capability that includes or is in communication a distributed network.
  • the data transmitted from the remote user 1202 may be training data, test data and/or live data to be processed by a learning machine.
  • the classifier is pre-trained so that the remote user sends live data.
  • the data transmitted by the customer is received at the central server 1206, which may transmit the data to one or more learning machines via an internal network 1214a-b.
  • learning machines may comprise SVMs, neural networks, random forests, Bayesian classifiers, or other learning machines or combinations thereof.
  • the web server 1206 is isolated from the learning machine(s) by way of a firewall 1208 or other security system.
  • the service provider 1212 may also be in communication with one or more financial institutions 1210, via the Internet 1204 or any dedicated or on- demand communications link.
  • the web server 1206 or other communications device may handle communications with the one or more financial institutions.
  • the financial institution(s) may comprise banks, Internet banks, clearing houses, credit or debit card companies, or the like. Where the remote user is using a smart phone, the financial institution may also be a wireless service provider.
  • the service provider may offer learning machine processing services via a web-site hosted at the web-server 1206 or another server in
  • a customer 1202 may transmit data to the web server 1206 to be processed by a learning machine.
  • the customer 1202 may also transmit identification information, such as a username, a password and/or a financial account identifier, to the web-server.
  • the web server 1206 may electronically withdraw a pre- determined amount of funds from a financial account maintained or authorized by the customer 1202 at a financial institution 1210.
  • the web server may transmit the customer's data to the classifier 1100.
  • the post-processed output is returned to the web-server 1206.
  • the output from a learning machine may be post-processed in order to generate a single-valued or multi-valued, computationally derived alpha-numerical classifier, for human or automated interpretation.
  • the web server 1206 may then ensure that payment from the customer has been secured before the post-processed output is transmitted back to the customer 1202 via the Internet 1204.
  • An exemplary implementation of the website may include five pages that hold the main functionalities: Welcome (which may serve as the index page); Facts:
  • the Statistics and Background information page may include introductory information about melanoma and why such an application may be beneficial to the user.
  • the MyLab page will be confidential to each patient user; Referrals: for directing patient users to medical professionals, e.g., physicians, dermatologists.
  • a navigation bar may be provided on each page for accessing the other pages.
  • the web server 420 should have scalable bandwidth. It should be noted that while the web server 420 and database server 430 are illustrated as physically separate servers, they may be combined within a single server.
  • the user's identification and geographical location may be stored on the database server 430 in the database of patients (users) and system administrators. Also stored on server 430 will be information about referral physicians and other relevant data. All three servers will be accessible to administrators.
  • administration interface will allow the administrators to perform the most basic tasks, which may include changing content in the web pages, deletion of users, or database entries.
  • the exemplary architecture described above combines a fully integrated service with a modular design that allows expansion and development of different modules, independent of the other modules.
  • the modular design will also permit implementation and maintenance of different modules/servers by different entities.
  • the database server may be a combination of different servers in which one server holds a referral database that is maintained by a for-fee referral service that contracts with physicians and receives compensation for each referral.
  • FIG. 6 illustrates the various components of the image pre-processing and feature selection operations.
  • the next stage of pre-processing block 214 includes execution of a segmentation algorithm (step 304) which is applied to the original image to isolate the area of interest from the background.
  • segmentation algorithm In biomedical image segmentation, most techniques can be categorized into three classes: (1) characteristic feature thresholding or clustering, (2) edge detection, and (3) region extraction tasks such as measurements and registration.
  • edge detection is used to identify the outer contours of the suspected lesion, as shown in FIG. 7. It will be readily apparent to those in the art that other segmentation methods may be used.
  • the image analysis server pre-processes the image by converting it into a gray scale image, smoothing and equalizing. This step facilitates extraction of the contour by setting a threshold on a histogram of the gray scale values. Different gray scale values may also be used to extract inner contours within the lesion.
  • a new feature set may include two types of features: geometrical features assessing the symmetry of the image and color spectrum features. Additional features may include shape spectral features.
  • a set of geometrical constructs may be used for the garbage image screening.
  • the gray level image obtained by averaging the RGB values was used and computed.
  • the standard deviation of the gray levels in five concentric rings shown in FIG. 19b was then calculated.
  • the minimum, maximum, and mean values of these five coefficients were used to evaluate three symmetry features.
  • a blob was assessed by correlating the image with nine masks featuring a Gaussian gray-level shape, as shown in FIG. 19c. The minimum, maximum, and mean value of the nine quantities were taken to produce three "blob" features.
  • classifiers were built with subsets of features: "Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob” (the three “Blob” (the three “blob” features), “Sym” (the three “Blob”
  • step 306 includes using the outer contour to extract geometrical features by first identifying the center of gravity, or centroid, of the image and determining the average radius. (See FIG. 8a.) Next, the difference in distance from the center or gravity to opposite points on the contour is determined. The coefficient of variance of the radius (“radius cv”) is determined by calculating the standard deviation of the distance of a contour point to the center of gravity minus radius, divided by the mean radius.
  • the mean radius may be determined from the image if the user took the photograph with some form of dimensional reference, such as a ruler or a coin with known dimensions, or the value may also be obtained from the user-entered responses to the survey questions if they have provided the dimensions of the suspected melanoma.
  • the diameter may be calculated using autofocus information contained within the EXIF meta data, if available.
  • the radius aspect is the ratio of the minimum radius to the maximum radius. Asymmetry is measured using the distances between opposite points on the contour and the center of gravity (dl and d2 in FIG. 8b). The average square difference between dl and d2 is computed for all contour points, then the square root is taken and divided by the mean radius to normalize.
  • island l_eccentricity is determined by measuring the distance between the center of gravity of the innermost contour and that of the outer contour, normalized by the mean radius.
  • the island2_eccentricity is the distance between the center of gravity of the second most inner contour and that of the outer contour, again normalized by the mean radius.
  • the contour length, "island l clength”, is the innermost contour normalized by the mean radius, and the island2_clength is the second most inner contour normalized by the mean radius.
  • step 308 color features are computed using the original, unpreprocessed image, but with the contours defined in the preceding process (step 304) superimposed over the original image.
  • Two sets of color features are used: the original RGB (red- blue-green) and the HSV (hue-saturation-value), providing a total of six channels.
  • each color channel is coded on 8 bits for a total of 24 bits. These values are converted to HSV using methods that are well known in the art.
  • the median value is computed within the contour and around the contour for both the outer and innermost contours using the narrow surrounding bands illustrated in FIGs. 9a and 9b for outer and inner contours, respectively.
  • a color feature in a given channel, except for hue, is defined as the relative difference between the inner and outer values:
  • Color feat 2 -(outer value - inner value)/ (outer value + inner value).
  • Hue (H) sin 27i(inner_value).
  • color R color G, color B, color H, color S, color V, and
  • color island R color island G, color island B, color island H,
  • color island S and color island V.
  • each of the features is standardized by subtracting the mean of all values of that feature and dividing the result by the standard deviation prior to loading into a feature matrix for classification.
  • the classifiers used in step 218 for analysis of the user-provided image and data are pre-trained, i.e., trained and tested, using one or more image datasets having known outcomes.
  • the feature matrix shown in FIG. 10 represents a training dataset with 103 cases of which 50 were images of malignant melanoma and 53 were controls including atypical nevi (27), atypical nevi demoscopy (7), congenital nevi (5), blue nevi (3), halo nevi (4), lentigo (6), and spitz nevus (1).
  • Table 2 lists the features in the order shown
  • the y-axis represents the 103 data samples, with the 53 controls at the top (#51-103) and 50 malignant samples (#1-50) at the bottom.
  • the right hand column of the table indicates the class (control vs. malignant).
  • An ensemble of classifiers was trained using the features extracted from the image, with one classifier trained on each of the feature types A, B, C and D.
  • an ensemble of classifiers is a set of classifiers whose individual decisions are combined in some way to classify new examples.
  • An ensemble may, but not necessarily, consist of a set of different classifier types. Table 3 lists the features that fall within the 4 feature types:
  • color V color R, color island G, color island B;
  • color island H color island S, color island V,
  • Each classifier of the ensemble uses standardized features, in which the mean of the features was subtracted and the result was divided by the standard deviation.
  • the normalization coefficients are computed on training data and the same values are applied to the test data.
  • each classifier in the ensemble is a support vector machine classifier, with a separate kernel (trained and tested) used for each of the feature types.
  • a radial basis function (RBF) or Gaussian kernel is used.
  • a linear SVM may also be used.
  • a univariate, Gaussian classifier may be used, however, other learning machine classifiers that are known in the art that may be used, including random forests, decision trees, neural networks and others, as well as combinations of different types of classifiers.
  • the output of each classifier may be postprocessed with linear logistic regression to obtain a mapping of outputs to probabilities.
  • FIG. 13 illustrates a basic example of a two level hierarchical system of classifiers, i.e., an ensemble of multiple classifiers that produces a final, integrated classification using a combination of input data types that are relevant to the knowledge to be discovered with base classifiers that process the different input data types for input into a second-level, integrating or stacking classifier.
  • one or more first-level, or base, classifiers 1302 A and 1302B may be trained and tested to process a first type of input data 1304 A, for example,
  • mammography data pertaining to a sample of medical patients.
  • One or more of these classifiers may comprise a distinct kernel.
  • Another one or more additional first-level classifiers 1302C and 1302D may be trained and tested to process a second type of data 1304B, for example, genomic data, for the same or a different sample of medical patients.
  • the additional classifiers may comprise a distinct kernel.
  • the output from each of the base classifiers may be compared with each other (i.e., output Al (1306A) compared with output A2 (1306B); output B1(1306C) compared with output B2(1306D) in order to determine optimal outputs (1308A and 1308B).
  • the optimal outputs from the two types of base classifiers 1308A and 1308B may be combined to form a new multi-dimensional input data set 1310, which in this example relates to combined mammography and genomic data.
  • the new data set may then be processed by one or more appropriately trained and tested second-level, or stacking, classifiers 1312A and 1312B.
  • the resulting outputs 1314A and 1314B from the second-level classifiers 1312A and 1312B may be compared to determine an optimal output 1316.
  • the optimal output 1316 may identify causal relationships between the mammography and genomic data points.
  • the contemplated hierarchy of learning machines may have applications in any field or industry in which analysis of data by a learning machine is desired.
  • the hierarchical processing of multiple data sets using multiple classifiers may be used as a method for pre-processing or post-processing data that is to be input to or output from still other learning machines.
  • pre-processing or post-processing of data may be performed to the input data and/or output of the above-described hierarchical architecture of learning machines.
  • FIG. 14 illustrates an ensemble or hierarchical approach applied to image analysis of a user-provided image 500.
  • the image analysis server 410 receives the downloaded image and performs the pre-processing steps 502A-502D needed to extract the ABCD features.
  • Each extracted feature is separately processed using a trained base classifier 504A-504D, respectively, that has been optimized for classification of the corresponding ABCD feature, to generate an output that identifies whether the characteristics of the features extracted from the image correspond to a diagnosis of melanoma.
  • the results of the classification of each of classifiers 504A-504D are combined to create the ensemble by overall classifier 506, generating a single result which is output for further processing.
  • the overall classifier 506 may be a trained learning machine, such as a second support vector machine, or may simply be a weighted combination of the results of the individual classifiers 504A-D.
  • the output of classifier 506 will typically be post-processed by the web server in step 508 to generate a result that is meaningful to the user.
  • the result that is provided to the sender/user is a risk score or probability, along with graphics and information that can assist in interpretation of the score.
  • the outputs of the individual classifiers 504A-D can be post-processed to generate a risk score or probability based on the corresponding A, B, C or D feature, which can also be provided to the sender/user with the overall risk score.
  • the overall classifier 506 operated by voting of the different classifiers 504A-D, with each individual classifier being given a weight of 1.
  • the resulting vote was postprocessed (step 508) with logistic regression to provide a probability value.
  • CLOP Challenge Learning Object Package
  • DD data ( ⁇ , ⁇ )
  • model_A chain( ⁇ standardize, fixed_fs(A), naive,klogistic ⁇ );
  • model_B chain( ⁇ standardize, fixed fs(B), naive,klogistic ⁇ );
  • model_C chain( ⁇ standardize, fixed_fs(C), naive,klogistic ⁇ );
  • model_D chain( ⁇ standardize, fixed fs(D), naive,klogistic ⁇ );
  • FIG. 11 shows the ROC curve for
  • FIG. 12 illustrates an exemplary format for a sample report to the user showing the result of classification of the image that is shown in FIG. 7.
  • the user name is provided along with a case number.
  • the photograph that was submitted by the user is reproduced as image 904.
  • the result of the ensemble classifier is displayed at 906 in the form of a confidence level that the suspected lesion is melanoma.
  • the classification is Melanoma with 96% confidence. This number is the post-processed logistic regression probability value of the ensemble classifier.
  • the lower portion of the figure shows the logistic regression probability values for the individual A, B and C classifiers as bar graphs 90A, 908B and 908C, respectively, showing the confidence level for the corresponding feature.
  • a D classifier was not used in this example because dimensional data was not available from the test image dataset that was used.
  • the bottom of the exemplary display window includes a statement describing the basis for the assessment and provides a recommendation that the user consult a physician if any of the features is symptomatic of melanoma with more than 50% confidence.
  • the message are the bottom of the display window may include a hyperlink to a different page that allows the user to enter their geographical location information to obtain a referral to one or more specialists who can further evaluate the suspected lesion.
  • the geographical location information has already been provided to the central server, selecting the hyperlink with direct the user to a page that automatically displays the names and contact information for one or more specialists.
  • Table 4 below lists the top ranking image features identified by the different feature selection methods with the top five of each method underlined:
  • D features were not included in the above test because dimensional data was not available from the images in the initial test database. (Note that, as discussed below, the dimensional data may be obtained from the survey responses.)
  • the three types of image features are selected in the top ranking features.
  • the A (asymmetry) features appears to be more important than the B (border) features.
  • the color features the color of the "island” appears to be a determining factor by way of its intensity relative to the surrounding area, its "redness", and its hue, which represents the color direction globally.
  • FIG. 15 provides a block diagram showing the modular and hierarchical arrangement of tasks for analyzing the combination of data obtained from the remote user of a smart phone or computer who wishes to obtain a preliminary screening for melanoma using the system and method of the present invention.
  • the user input 600 includes the photograph(s) of the suspected melanoma sites, the survey data, the geographical location information and the (optional) account information for securing payment.
  • the input 600 is downloaded to the image server for processing of the image and survey data in processing modules 606 and 604, respectively.
  • the image analysis module 606 as illustrated includes manual image pre-processing 620, which is shown in dashed lines because it is actually performed by the user under instruction from the image server, and is not part of the image server process.
  • the manually pre-processed image is segmented in step 622 after which the border and geometric parameters are extracted in step 630. These values are separated into A (asymmetry) and B (border) groups and input into their respective feature classifiers, Classifier A (632) and Classifier B (634). Using the borders detected in the segmentation step, the color features are extracted from the original (manually pre-processed) image in step 624 and input into Classifier C (626). If the dimensional information can be obtained from the image using the EXIF meta data, or if the user included in the photograph a
  • Classifier D is only indicated by dashed lines within image analysis module 606 because it may not be possible to obtain the dimensional data from the image. In this case, the dimensional data will be obtained through the survey data that is input into the survey processing module 604 of the image analysis server.
  • Survey processing module 604 receives the responses to the survey data.
  • This mole or freckle is located on my
  • SI 7 Geographic My residence city & state or mail (zip) code is . location
  • a threshold of 6 to 7 mm may be used to diagnose malignant melanoma based on size of the mole.
  • Classifier D would be a learning machine such as a support vector machine, random forest, linear logistic regression model (sigmoid) or other classifier that is trained with data having known outcomes.
  • sigmoid linear logistic regression model
  • a sigmoid function is used for determining probability of cancer based on the relationship:
  • FIG. 16 illustrates the exemplary sigmoid function.
  • the results of each of Classifiers A, B, C and D will be input into ensemble classifier 640 to generate an overall result based on analysis of the image. This result will preferably be converted into a probability or percentage to provide the user with a risk score.
  • the survey answers will be extracted from the user survey response in the feature extraction process step 612 and input into a trained classifier (Classifier E (616)), which, as in the other classifiers, may be a support vector machine (SVM), random forest (RF), or other learning machine as is known in the art.
  • Classifier E 616)
  • SVM support vector machine
  • RF random forest
  • the survey features SI through S16 in the example shown in Table 5 are all binary.
  • some or all of the survey questions can request an explanatory answer, with the survey providing a suggested format or multiple-choice options for the answer.
  • the survey could provide a list of options such as torso, head, neck, face, etc., possibly associated with a number, so that the user would enter either the word “neck” or the number, e.g., "3", corresponding to "neck.”
  • options such as torso, head, neck, face, etc., possibly associated with a number, so that the user would enter either the word “neck” or the number, e.g., "3", corresponding to "neck.”
  • SVM trained classifier
  • FIG. 18 illustrates an example of such a survey.
  • the display includes an image of the lesion 1802 so that the survey response can be matched up with the particular lesion of interest.
  • elements A, B and C of the survey may be automatically entered by the system based on the values extracted during the pre-processing of the image that was downloaded prior to completion of the survey.
  • diameter D and evolution E may be selected from within a range of no change to rapid change.
  • This approach is ideal for use with a smart phone with a touch screen, where the user can select the value using a slider activated by dragging a finger across the screen.
  • diameter D the user moves slider 1816 until the appropriate numerical value is displayed to the right of the slider.
  • evolution E the user moves slider 1818 until the appropriate descriptive term appears next to the slider.
  • help screens may be provided to assist the user in selecting appropriate values for A, B and C.
  • a help screen for asymmetry can display: "A is for Asymmetry:
  • a mole is considered symmetric if it is roundly shaped.
  • a mole may be oval, which is one kind of mild asymmetry. It can have no axis of symmetry at all.”
  • Sample images of asymmetric moles may be included.
  • An example of a help screen for border is "B is for Border: An irregular border is an indication of malignancy, except for very small moles, which may have an irregular border due to the skin texture. Border distortions occur in pictures taken with insufficient light; take pictures outside with indirect light.”
  • a help screen for color may provide examples of different color ranges, while a diameter help screen may include labeled circles of different to allow the user to compare the mole to the circles to select the closest size circle.
  • the probability of cancer given the image evidence (or the survey for D) for the ABCD features can be obtained from a logistic regression model or can be combined similarly to the above equation.
  • S I , S2 and S3 are major indicators of melanoma.
  • the major indicators of changes in size, shape and color of a new or pre-existing cutaneous lesion were seen in 94%, 95% and 89%>, respectively, of the lesions evaluated.
  • a value of ⁇ was selected somewhat arbitrarily as 0.45 for the major signs.
  • Each column of Table 5, except for the last column, represents the probability of cancer given the single feature evidence for A, B and C obtained from FIG. 7 and the survey responses. For simplicity, survey responses are for S I - S8 only. These partial results are combined with the noisy-OR model of Equation 1 to produce the result provided in the last column.
  • the rows represent three hypothetical mole diagnoses based on different possible survey responses in combination with the image of FIG. 7.
  • Equation 1 Even with no survey results, the probability (risk score) is higher than that determined by ABCD classifier 640 using the image data alone, which had diagnosed a 96% chance of malignancy. Thus, the model of Equation 1 appears to be overly sensitive, possibly due to an assumption of independence between variables. It is a known advantage of SVMs that they do not make independence assumptions. Thus, a better result can be obtained when ABCDE classifier 642 is SVM-based. In a preferred embodiment, both ensemble classifiers 640 and 642 will be SVMs.
  • the input from user 600 is also provided to the web server module 608, which extracts user information, including the geographical location in step 648.
  • the geographical location can be automatically extracted from GPS information available from a smart phone with a GPS function enabled, or from EXIF data, if available.
  • the geographical location is compared to a database containing physician or other healthcare provider information to identify practitioners within a certain distance of the user.
  • the referral match-up operation 652 identifies providers who match the geographical location of the user and forwards the contact information and, preferably, distance information from the user's location to the provider's office, to be output to the remote user in block 670.
  • the information that is provided to the remote user is a list of physicians' names, their respective addresses, telephone numbers and distances, in miles or kilometers, from the user's geographical location.
  • Optional transactional module 610 receives information from the user 600 that allows a transaction to be conducted to secure payment for the analysis services.
  • User information 600 will include a financial account number, which may be a credit card number, PayPal ® account number, wireless service account number or other account to which a charge can be submitted and payment received.
  • This account information will be communicated to a financial institution, e.g., bank, credit card company, PayPal , wireless provider, etc., for entry into its database 622 to show a charge against the user's.
  • the charge per transaction will be on the order of $2 to $20, however, provisions may be made to make a one-time payment to establish an account with the analysis service provider to allow the user to submit a fixed or unlimited number of inquiries.
  • the account information that is transmitted will be the user's account with the analysis service provider.
  • the web server will compare that account information with its own database to confirm that the user has an account that has been paid up or will be billed for the service. If the user's account has expired or is otherwise unavailable, the server will notify the user and no analysis will be performed until arrangements for payment have been made.
  • the user information provided to the transactional module 610 will be the user's geographical information which will be compared to information in the referral database 650.
  • the provider(s) who would receive a referral based on a match-up of user's geographical location would have their own financial account charged for the analysis service.
  • the analysis service would be an advertising or marketing expense for the provider.
  • the provider could be the physician who would treat the user, or it could be a diagnostic laboratory that has contracted with one or more physicians to whom the user would be referred under the assumption that the physician would send pathology samples to the laboratory for analysis, for which the laboratory would be compensated by the patient or the patient's insurance.
  • the results of the analysis plus other useful information for interpretation of the result will be transmitted to the user in block 670 along with the referral information for healthcare providers that are identified in the referral match-up 652.
  • FIGs. 17a-17d illustrate exemplary content that can be displayed on a smart phone such as an iPhone ® or similar touch screen Internet-capable phone.
  • the smart phone implementation may have several main functionalities.
  • a welcome page will be displayed when the smart phone user initially accesses the application.
  • the welcome page will include basic information about the service and will have icons for selecting different pages. In one embodiment, if no icon is "clicked", after 3 seconds, the application will switch automatically to the camera page.
  • the "camera" page 1600 allows the user to take photographs.
  • the functionality of the camera software of the application will include allowing the user to zoom and take pictures using auto focus.
  • the goal is to obtain an image as close as possible to raw data, preferably in a non-compressed format.
  • the auto focus information will be records to allow the image processor to determine the distance of the camera from the lesion.
  • the zooming factor should also be recorded.
  • feedback may be given to the user to allow him or her to take better pictures.
  • the feedback may include illumination ("is there sufficient light?"), jitter and blur (“is the camera or the body part moving?”), focus (“is the camera far enough from the lesion to allow the auto-focus to work properly?”), and framing (“is the mole well- framed in the square?”, referring to a square displayed at the center of the screen to assist in centering and zooming the image).
  • Sample instructions may include "Use outside indirect light. Rest camera on support 3 inches (8 cm) away from mole. Zoom to fit framing rectangle (or center mole and zoom to max)"
  • a "help” button 1604 can be pressed to obtain more detailed instructions.
  • a slider 1606 allows the user to zoom the camera to fit the frame 1610, which is displayed on the screen. When the user is ready, he or she clicks on the camera button 1608 to take the photograph. Once the picture has been taken, the user is automatically sent to the "album" page 1620, shown in FIG. 17b. Alternatively, the user can click on the album icon 1612 at the bottom of the page.
  • Album page 1620 displays images 1622, 1624 and 1626 that have already been take by the user.
  • a ticker or banner at the top of the page instructs the user to "Select the picture you would like to send for diagnosis, or go back to the camera to take another picture.” If the user wishes to take another photograph, he or she will click on the camera icon 1614. If the user is satisfied with the photographs and has taken photographs of all of the suspicious moles, the user clicks on one of the images to be automatically directed to the "send" page.
  • the send page (not shown) allows the user to zoom the image, delete the image, go back to the album page 1620, or send the image to the image analysis server for analysis after responding to the survey questions that are displayed on the send page. After completion of the survey and selection of one or more image, the user can click on the send icon 1616 to transmit the photograph and survey data to the image analysis server.
  • the display will automatically change to the results page 1640, shown in FIG. 17c.
  • the ticker or box at the top may display a message, "These are your results. Check periodically for updates".
  • a table 1642 is shown with rows corresponding to three different images.
  • the second column 1644, first row displays a set of bar graphs corresponding to the classification results of the individual ABC feature classifiers, such as the example shown in FIG. 12.
  • the third column 1646 provides the risk score which, for the first image, is a 96% probability of melanoma.
  • the second and third rows are shown as being "In progress" because the image analysis server is still analyzing the images. Once the analysis if completed, the table will be updated and the results will be displayed in the same format as shown in the first row of the table.
  • the user can obtain referrals to physicians by pressing the physician icon 1648 from any of the other pages.
  • the referral page 1660 shown in FIG. 17d, may display one or more (if available) physicians in the area, providing a table with the name, address and telephone number of the physician, an optional photograph, if available, and the approximate distance from the user's location to the physician's office.
  • the physician information may include a live link that will automatically dial the physician's office telephone number when the user clicks on the contact information in the second column or on a separate "call" button that may be displayed as a fourth column (not shown) in the table.
  • Example 1 Classifier variation
  • the ABC features often fail because of poor segmentation because the mole outline cannot be found. Consequently, the features designed for separating garbage from skin disease, which involve no detection of mole outline, work as well or better.
  • Adding yet more features may help. For example, a number of mistakes are due to the presence of hairs. Thus, hairs may need to be removed, or a texture analysis may be appropriate.
  • the linear kernel recognizer does not overfit data.
  • the AUC obtained on training data is 0.8.
  • the ABC and garbage image recognizers were integrated into a single system and combined with the D and E features.
  • the user is provided ABCDE scores and an overall risk assessment: Low, Medium, or High.
  • the "garbage” classifier identifies a "garbage” image.
  • the color feature C is lower than 0.2.
  • the area of the mole relatively to the area of a round mole of the same diameter is lower than 0.5 or larger than 1.2.
  • the A, B, and C recognizers trained with the CLOP models from subsets of lower level features produce the A, B, and C feature scores:
  • model_A chain( ⁇ standardize, fixed_fs(A), naive, klogistic ⁇ );
  • model_B chain( ⁇ standardize, fixed fs(B), naive, klogistic ⁇ );
  • model_C chain( ⁇ standardize, fixed fs(C), naive, klogistic ⁇ );
  • the D feature is obtained from the diameter d using a squashing function:
  • the E feature was the average of survey.recent
  • the survey may ask for speed of evolution, which is a metric frequently used by
  • ABCDE_risk min(ABCD_risk+l , 2) if E_risk>0
  • Tables 7 and 8 show ABC features only and ABCDE features, respectively. There were no false negatives (melanoma cases found low risk) and a few false positives. Adding the D and E features improves the results, but not significantly. The fraction of rejected pictures is rather high due to partly to the low quality of the images and the large number of non-mole images (that are correctly rejected since the classifier focuses on evaluating the risk of melanoma for moles.
  • the DERMATLAS image dataset from Johns Hopkins University (available on the World Wide Web at dermatlas.org) was used to extract 1000 images of skin diseases with various diagnoses, including herpes, acnea, drug reactions, insect bites, squamous cell carcinoma, basal cell carcinoma, congenital nevi, other benign moles, and melanoma.
  • Associate with the images are annotations both in the form of MS access database and MS Excel spreadsheets.
  • the annotations include age and gender of the patient, data source information, and diagnosis.
  • the software was written to crop the images into 1717 cropped pictures (some coming from the same image).
  • the software also allows a framing box to be drawn around the lesion, such as described above with reference to FIG. 17a.
  • the framing box facilitates segmentation of the mole from the background and is in a similar form to the data that would be collected using the smart phone application.
  • the original annotations do not include a database entry for lesion size or diameter and for lesion evolution, which are needed for accurate classification.
  • the database includes comments that often have an indication of size and/or evolution.
  • the non-cropped image permits determination of the context and evaluation of the scale.
  • a data browser was written to visualize the data and the annotations, and to allow editing of the annotations, thus permitting the extraction of approximate values of size and evolution for every cropped image. Such values are not very accurate, but, they provided a significant improvement in classification
  • Fast The word “fast” appears in relation to growth description, or it appears from the description that there has been a fast evolution in the last few months. If the diagnosis is that of an infectious disease or drug eruption or insect bite, but there is no mention of evolution, it is also categorized as fast.
  • Image segmentation is used to isolate the mole from its background. This tends to be a difficult step because of the high variability of the images.
  • the sequence below was used to find a mask outlining the mole.
  • the search for a lesion contour was limited to an area slightly larger than the framing box (a border of 1/5 of the width or height is added around the framing box). If several lesions were present in the area of interest, only the largest lesion was retained.
  • Suspicious contour the lesion contour is not well centered in the framing square, too small or too large compared to the green square, or partly outside the framing square.
  • FIG. 20 is a flow diagram of the ABCDE risk analysis system in which the image is provided by a user with an iPhone ® .
  • a classifier was not trained in all features because there were few cases of melanoma in the database— only about 30 of which are "typical" cases that can be classified with this method (the other melanoma cases include ulcers, metastases and nails). Instead, the data was used for testing only, and a heuristic formula was used to build an ABCDE classifier from the ABC classifier and the D and E features:
  • ABC is the output of the ABC classifier
  • D is obtained from the diameter (after passing it through a squashing function to normalize it between 0 and 1)
  • E is 0.1 for non-evolving moles, 0.5 for slow evolution, and 0.9 for fast evolution.
  • a larger weight was assigned to the E feature to give a similar importance to the geometrical features ABCD and the evolution E.
  • the index ABCDE was normalized between 0 and 1 and two thresholds were set:
  • Borderline Color not suspicious, but ABCDE index indicates medium of high risk.
  • the "Borderline” category was introduced in an effort to reduce the number of false positives and false negatives. It plays the role of a buffer like the “medium” risk category, but is based on a different criterion. While this may be a useful distinction, it can be lumped with the “medium” category. Similarly, “Low contrast” could be lumped with “No call”, but it falls in between “Low risk” and "No call”.
  • Tables 9 and 10 show the number of examples in a confusion matrix. Truth values are shown horizontally and risk assessment vertically. Comparing the old (Table 9) and the new preprocessing (Table 10), the new preprocessing decreases the number of "No call” classifications without increasing significantly the number of errors, and false negative are reduced. The answers were rated as very good, acceptable and bad, which are indicated in the tables as "A+", "A" and "B", respectively.
  • Segmentation error part of the mole is not detected because there are either several clusters or large differences in color between regions: 14%
  • Glare/shades/poor light 9%
  • the inventive system gives a recommendation or comment, which provides more details about how the decision was made and whether to consult a doctor. With this comment, a meaningful assessment can be provided, even in the case of the three "no call" cases (Borderline, Low contrast, and No call).
  • the Table 11 provides a few examples of comments that might be provided:
  • the system and method of the present invention provide a free or low cost preliminary skin cancer screening capability that is accessible to the average person with a smart phone or a digital camera and Internet access.
  • the analytical services provided according to the invention are not intended to replace evaluation and diagnosis by a physician specializing in skin cancer, but are merely intended to assist an individual to determine whether they should see a physician for evaluation of an area of the skin that is of concern.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Surgery (AREA)
  • Pathology (AREA)
  • Physics & Mathematics (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Artificial Intelligence (AREA)
  • Urology & Nephrology (AREA)
  • Business, Economics & Management (AREA)
  • General Business, Economics & Management (AREA)
  • Databases & Information Systems (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Immunology (AREA)
  • Data Mining & Analysis (AREA)
  • Vascular Medicine (AREA)
  • Evolutionary Computation (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Endocrinology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Fuzzy Systems (AREA)
  • Multimedia (AREA)
  • Dermatology (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)
  • Image Analysis (AREA)

Abstract

La présente invention concerne un système et un procédé de diagnostic de mélanome à partir d'images numériques prises par un utilisateur distant avec un téléphone intelligent ou une caméra numérique et transmises à un serveur d'analyse d'images en communication avec un réseau distribué. Le serveur d'analyse d'images comprend une machine d'apprentissage programmée pour la classification d'images de lésions cutanées malignes et bénignes. L'image fournie par l'utilisateur est prétraitée pour extraire les caractéristiques de dimension, de forme et de couleur lors du traitement par la machine d'apprentissage programmée pour classer la lésion suspectée. Le résultat de la classification est post-traité pour générer une note de risque qui est transmise à l'utilisateur distant. Une base de données associée au serveur comprend des informations de référence pour l'association géographique de l'utilisateur distant à un médecin local pour le traitement du cancer de la peau et pour fournir les informations de contact à l'utilisateur. Une opération facultative comprend la collecte d'informations financières pour sécuriser le paiement des services d'analyse.
PCT/US2010/061667 2009-12-22 2010-12-21 Système et procédé de dépistage de mélanome à distance Ceased WO2011087807A2 (fr)

Applications Claiming Priority (8)

Application Number Priority Date Filing Date Title
US28937209P 2009-12-22 2009-12-22
US61/289,372 2009-12-22
US29398710P 2010-01-11 2010-01-11
US61/293,987 2010-01-11
US30479610P 2010-02-15 2010-02-15
US61/304,796 2010-02-15
US30879210P 2010-02-26 2010-02-26
US61/308,792 2010-02-26

Publications (2)

Publication Number Publication Date
WO2011087807A2 true WO2011087807A2 (fr) 2011-07-21
WO2011087807A3 WO2011087807A3 (fr) 2011-11-17

Family

ID=44304894

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2010/061667 Ceased WO2011087807A2 (fr) 2009-12-22 2010-12-21 Système et procédé de dépistage de mélanome à distance

Country Status (1)

Country Link
WO (1) WO2011087807A2 (fr)

Cited By (42)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014509874A (ja) * 2010-10-11 2014-04-24 ウーンドマトリックス、インク． 創傷管理モバイル画像キャプチャデバイス
WO2015013288A2 (fr) 2013-07-22 2015-01-29 The Rockefeller University Système et procédé de détection optique de maladies de la peau
WO2016144239A1 (fr) * 2015-03-06 2016-09-15 Instant Advice Ab Procédé et système de fourniture de conseil médical concernant un traitement d'un état de santé d'un utilisateur
WO2016210369A1 (fr) * 2015-06-26 2016-12-29 Kci Licensing, Inc. Système et procédés de mise en place de protocoles de traitement de plaie
EP3054846A4 (fr) * 2013-10-08 2017-10-18 Leaf Healthcare, Inc. Procédés, dispositifs et techniques de détection d'escarres
WO2018005316A1 (fr) 2016-07-01 2018-01-04 Bostel Technologies, Llc Phonodermoscopie, système et procédé dispositif médical destiné au diagnostic de la peau
US10004447B2 (en) 2010-04-22 2018-06-26 Leaf Healthcare, Inc. Systems and methods for collecting and displaying user orientation information on a user-worn sensor device
IT201600132357A1 (it) * 2016-12-29 2018-06-29 Univ Degli Studi Padova Metodo e apparecchiatura di mappatura tridimensionale di una porzione della cute di un paziente
KR101886990B1 (ko) * 2016-12-02 2018-08-08 아벤트, 인크. 의료적 이미징 기반 절차에서 타겟 해부 대상으로의 내비게이션을 위한 시스템 및 방법
US10140837B2 (en) 2010-04-22 2018-11-27 Leaf Healthcare, Inc. Systems, devices and methods for the prevention and treatment of pressure ulcers, bed exits, falls, and other conditions
CN109191407A (zh) * 2018-09-20 2019-01-11 湘潭大学 一种基于极限学习机的碎纸片拼接复原方法及系统
WO2019027667A1 (fr) * 2017-08-01 2019-02-07 The General Hospital Corporation Système et procédé d'analyse prédictive de lésions mammaires pathologiques
US10258258B2 (en) 2010-03-07 2019-04-16 Leaf Healthcare, Inc. Systems, devices and methods for the prevention and treatment of pressure ulcers, bed exits, falls, and other conditions
DE102017010397A1 (de) 2017-11-09 2019-05-09 FotoFinder Systems GmbH Verfahren zur Auswertung des Hautbildes eines Menschen im Rahmen der Hautkrebs-Vorsorgeuntersuchung sowie Vorrichtung zu dessen Betrieb
RU2688971C2 (ru) * 2014-05-15 2019-05-23 Колопласт А/С Способ и устройство для захвата и сохранения в цифровой форме изображений места раны, свища или стомы
AT16426U1 (de) * 2016-03-02 2019-08-15 Mathias Zirm Dr Ferndiagnose-Unterstützungsverfahren
CN110195094A (zh) * 2019-05-30 2019-09-03 湖南自兴智慧医疗科技有限公司 一种远程染色体核型检测方法和系统
US10499845B2 (en) 2014-08-25 2019-12-10 Singapore University Of Technology And Design Method and device for analysing an image
US10588565B2 (en) 2010-04-22 2020-03-17 Leaf Healthcare, Inc. Calibrated systems, devices and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US10595805B2 (en) 2014-06-27 2020-03-24 Sunnybrook Research Institute Systems and methods for generating an imaging biomarker that indicates detectability of conspicuity of lesions in a mammographic image
US10631732B2 (en) 2009-03-24 2020-04-28 Leaf Healthcare, Inc. Systems and methods for displaying sensor-based user orientation information
WO2020113326A1 (fr) * 2018-12-04 2020-06-11 Jiang Ruowei Diagnostics cutanés automatiques basés sur l'image à l'aide d'un apprentissage profond
CN111275703A (zh) * 2020-02-27 2020-06-12 腾讯科技（深圳）有限公司 图像检测方法、装置、计算机设备和存储介质
CN111310841A (zh) * 2020-02-24 2020-06-19 中南大学湘雅医院 医学图像分类方法、装置、设备、计算机设备和存储介质
US10758162B2 (en) 2010-04-22 2020-09-01 Leaf Healthcare, Inc. Systems, devices and methods for analyzing a person status based at least on a detected orientation of the person
US11051751B2 (en) 2010-04-22 2021-07-06 Leaf Healthcare, Inc. Calibrated systems, devices and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US11134885B2 (en) 2015-08-13 2021-10-05 The Rockefeller University Quantitative dermoscopic melanoma screening
US11272860B2 (en) 2010-04-22 2022-03-15 Leaf Healthcare, Inc. Sensor device with a selectively activatable display
US11278237B2 (en) 2010-04-22 2022-03-22 Leaf Healthcare, Inc. Devices, systems, and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US11298072B2 (en) 2016-07-01 2022-04-12 Bostel Technologies, Llc Dermoscopy diagnosis of cancerous lesions utilizing dual deep learning algorithms via visual and audio (sonification) outputs
CN114451870A (zh) * 2022-04-12 2022-05-10 中南大学湘雅医院 色素痣恶变风险监测系统
US11369309B2 (en) 2010-04-22 2022-06-28 Leaf Healthcare, Inc. Systems and methods for managing a position management protocol based on detected inclination angle of a person
CN115082691A (zh) * 2022-07-28 2022-09-20 吾征智能技术(北京)有限公司 基于深度学习的黑色素瘤识别方法及装置
US20230081581A1 (en) * 2020-01-30 2023-03-16 Termo Health Tecnologia Ltda Mobile system and auxiliary method for evaluating thermographic breast images
CN116705325A (zh) * 2023-06-26 2023-09-05 国家康复辅具研究中心 一种伤口感染风险评估方法及其系统
US11908154B2 (en) 2021-02-04 2024-02-20 Fibonacci Phyllotaxis Inc. System and method for evaluating tumor stability
US11980449B2 (en) 2010-04-22 2024-05-14 Leaf Healthcare, Inc. Systems and methods for monitoring orientation and biometric data using acceleration data
WO2024104704A1 (fr) * 2022-11-15 2024-05-23 Biotronik Ag Système fondé sur l'ia pour la surveillance de patients souffrant d'une maladie vasculaire périphérique
USD1036665S1 (en) 2021-02-04 2024-07-23 Fibonacci Phyllotaxis Inc. Transparent tumor size measuring ruler with transparent slider
USD1041654S1 (en) 2021-02-04 2024-09-10 Fibonacci Phyllotaxis Inc. Transparent tumor size measuring ruler
GB2632164A (en) * 2023-07-27 2025-01-29 Boots Co Plc Radiance measurement method and system
US12327633B2 (en) 2014-10-14 2025-06-10 Solventum Intellectual Properties Company System for monitoring compliant usage of negative pressure wound therapy

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596434B (zh) * 2018-03-23 2019-08-02 卫盈联信息技术（深圳）有限公司 欺诈检测和风险评估方法、系统、设备及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4437202B2 (ja) * 2004-01-09 2010-03-24 学校法人慶應義塾 色素沈着部位の遠隔診療システム
US8109875B2 (en) * 2007-01-03 2012-02-07 Gizewski Theodore M Derma diagnostic and automated data analysis system
US8213695B2 (en) * 2007-03-07 2012-07-03 University Of Houston Device and software for screening the skin

Cited By (71)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10631732B2 (en) 2009-03-24 2020-04-28 Leaf Healthcare, Inc. Systems and methods for displaying sensor-based user orientation information
US10874330B2 (en) 2010-03-07 2020-12-29 Leaf Healthcare, Inc. Systems, devices and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US10682076B2 (en) 2010-03-07 2020-06-16 Leaf Healthcare, Inc. Systems and methods for monitoring the attachment and/or positioning of a wearable of a sensor device
US10258258B2 (en) 2010-03-07 2019-04-16 Leaf Healthcare, Inc. Systems, devices and methods for the prevention and treatment of pressure ulcers, bed exits, falls, and other conditions
US11051751B2 (en) 2010-04-22 2021-07-06 Leaf Healthcare, Inc. Calibrated systems, devices and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US10888251B2 (en) 2010-04-22 2021-01-12 Leaf Healthcare, Inc. Systems, devices and methods for analyzing the attachment of a wearable sensor device on a user
US11317830B2 (en) 2010-04-22 2022-05-03 Leaf Healthcare, Inc. Systems and methods for managing pressurization timers for monitoring and/or managing a person's position
US11278237B2 (en) 2010-04-22 2022-03-22 Leaf Healthcare, Inc. Devices, systems, and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US10004447B2 (en) 2010-04-22 2018-06-26 Leaf Healthcare, Inc. Systems and methods for collecting and displaying user orientation information on a user-worn sensor device
US11272860B2 (en) 2010-04-22 2022-03-15 Leaf Healthcare, Inc. Sensor device with a selectively activatable display
US10588565B2 (en) 2010-04-22 2020-03-17 Leaf Healthcare, Inc. Calibrated systems, devices and methods for preventing, detecting, and treating pressure-induced ischemia, pressure ulcers, and other conditions
US10912491B2 (en) 2010-04-22 2021-02-09 Leaf Healthcare, Inc. Systems, devices and methods for managing pressurization timers for monitoring and/or managing a person's position
US10140837B2 (en) 2010-04-22 2018-11-27 Leaf Healthcare, Inc. Systems, devices and methods for the prevention and treatment of pressure ulcers, bed exits, falls, and other conditions
US11369309B2 (en) 2010-04-22 2022-06-28 Leaf Healthcare, Inc. Systems and methods for managing a position management protocol based on detected inclination angle of a person
US11883154B2 (en) 2010-04-22 2024-01-30 Leaf Healthcare, Inc. Systems and methods for monitoring a person's position
US10758162B2 (en) 2010-04-22 2020-09-01 Leaf Healthcare, Inc. Systems, devices and methods for analyzing a person status based at least on a detected orientation of the person
US12329536B2 (en) 2010-04-22 2025-06-17 Leaf Healthcare, Inc. Systems and methods for monitoring pressurization related information for managing a person's position
US10729357B2 (en) 2010-04-22 2020-08-04 Leaf Healthcare, Inc. Systems and methods for generating and/or adjusting a repositioning schedule for a person
US11948681B2 (en) 2010-04-22 2024-04-02 Leaf Healthcare, Inc. Wearable sensor device and methods for analyzing a persons orientation and biometric data
US11980449B2 (en) 2010-04-22 2024-05-14 Leaf Healthcare, Inc. Systems and methods for monitoring orientation and biometric data using acceleration data
JP2014509874A (ja) * 2010-10-11 2014-04-24 ウーンドマトリックス、インク． 創傷管理モバイル画像キャプチャデバイス
AU2014293317B2 (en) * 2013-07-22 2019-02-14 The Rockefeller University Optical detection of skin disease
JP2019058681A (ja) * 2013-07-22 2019-04-18 ザ ロックフェラー ユニバーシティーＴｈｅ Ｒｏｃｋｅｆｅｌｌｅｒ Ｕｎｉｖｅｒｓｉｔｙ 皮膚疾患の光学検出のためのシステム及び方法
US10307098B2 (en) 2013-07-22 2019-06-04 The Rockefeller University System and method for optical detection of skin disease
US10182757B2 (en) 2013-07-22 2019-01-22 The Rockefeller University System and method for optical detection of skin disease
WO2015013288A2 (fr) 2013-07-22 2015-01-29 The Rockefeller University Système et procédé de détection optique de maladies de la peau
EP3024383A4 (fr) * 2013-07-22 2017-04-05 The Rockefeller University Système et procédé de détection optique de maladies de la peau
JP2016530917A (ja) * 2013-07-22 2016-10-06 ザ ロックフェラー ユニバーシティーＴｈｅ Ｒｏｃｋｅｆｅｌｌｅｒ Ｕｎｉｖｅｒｓｉｔｙ 皮膚疾患の光学検出のためのシステム及び方法
US11931164B2 (en) 2013-07-22 2024-03-19 The Rockefeller University System and method for optical detection of skin disease
EP3054846A4 (fr) * 2013-10-08 2017-10-18 Leaf Healthcare, Inc. Procédés, dispositifs et techniques de détection d'escarres
RU2688971C2 (ru) * 2014-05-15 2019-05-23 Колопласт А/С Способ и устройство для захвата и сохранения в цифровой форме изображений места раны, свища или стомы
US11389108B2 (en) 2014-05-15 2022-07-19 Coloplast A/S Method and device for capturing and digitally storing images of a wound, fistula or stoma site
US10595805B2 (en) 2014-06-27 2020-03-24 Sunnybrook Research Institute Systems and methods for generating an imaging biomarker that indicates detectability of conspicuity of lesions in a mammographic image
US10499845B2 (en) 2014-08-25 2019-12-10 Singapore University Of Technology And Design Method and device for analysing an image
US12327633B2 (en) 2014-10-14 2025-06-10 Solventum Intellectual Properties Company System for monitoring compliant usage of negative pressure wound therapy
WO2016144239A1 (fr) * 2015-03-06 2016-09-15 Instant Advice Ab Procédé et système de fourniture de conseil médical concernant un traitement d'un état de santé d'un utilisateur
WO2016210369A1 (fr) * 2015-06-26 2016-12-29 Kci Licensing, Inc. Système et procédés de mise en place de protocoles de traitement de plaie
US11134885B2 (en) 2015-08-13 2021-10-05 The Rockefeller University Quantitative dermoscopic melanoma screening
AT16426U1 (de) * 2016-03-02 2019-08-15 Mathias Zirm Dr Ferndiagnose-Unterstützungsverfahren
WO2018005316A1 (fr) 2016-07-01 2018-01-04 Bostel Technologies, Llc Phonodermoscopie, système et procédé dispositif médical destiné au diagnostic de la peau
US11484247B2 (en) 2016-07-01 2022-11-01 Bostel Technologies, Llc Phonodermoscopy, a medical device system and method for skin diagnosis
US11298072B2 (en) 2016-07-01 2022-04-12 Bostel Technologies, Llc Dermoscopy diagnosis of cancerous lesions utilizing dual deep learning algorithms via visual and audio (sonification) outputs
EP3478160A4 (fr) * 2016-07-01 2020-04-08 Bostel Technologies, LLC Phonodermoscopie, système et procédé dispositif médical destiné au diagnostic de la peau
KR101886990B1 (ko) * 2016-12-02 2018-08-08 아벤트, 인크. 의료적 이미징 기반 절차에서 타겟 해부 대상으로의 내비게이션을 위한 시스템 및 방법
KR101949114B1 (ko) 2016-12-02 2019-02-15 아벤트, 인크. 의료적 이미징 기반 절차에서 타겟 해부 대상으로의 내비게이션을 위한 시스템 및 방법
WO2018122793A1 (fr) * 2016-12-29 2018-07-05 Universita' Degli Studi Di Padova Procédé et dispositif de cartographie tridimensionnelle de la peau d'un patient pour la prise en charge du diagnostic du mélanome
IT201600132357A1 (it) * 2016-12-29 2018-06-29 Univ Degli Studi Padova Metodo e apparecchiatura di mappatura tridimensionale di una porzione della cute di un paziente
WO2019027667A1 (fr) * 2017-08-01 2019-02-07 The General Hospital Corporation Système et procédé d'analyse prédictive de lésions mammaires pathologiques
DE102017010397A1 (de) 2017-11-09 2019-05-09 FotoFinder Systems GmbH Verfahren zur Auswertung des Hautbildes eines Menschen im Rahmen der Hautkrebs-Vorsorgeuntersuchung sowie Vorrichtung zu dessen Betrieb
WO2019091509A1 (fr) 2017-11-09 2019-05-16 FotoFinder Systems GmbH Procédé d'évaluation de la mesure de l'image de la peau d'une personne dans le contexte de l'examen de dépistage de cancer de la peau et dispositif de mise en œuvre de ce procédé
CN109191407A (zh) * 2018-09-20 2019-01-11 湘潭大学 一种基于极限学习机的碎纸片拼接复原方法及系统
US11553872B2 (en) 2018-12-04 2023-01-17 L'oreal Automatic image-based skin diagnostics using deep learning
CN114502061B (zh) * 2018-12-04 2024-05-28 巴黎欧莱雅 使用深度学习的基于图像的自动皮肤诊断
CN114502061A (zh) * 2018-12-04 2022-05-13 巴黎欧莱雅 使用深度学习的基于图像的自动皮肤诊断
US11832958B2 (en) 2018-12-04 2023-12-05 L'oreal Automatic image-based skin diagnostics using deep learning
WO2020113326A1 (fr) * 2018-12-04 2020-06-11 Jiang Ruowei Diagnostics cutanés automatiques basés sur l'image à l'aide d'un apprentissage profond
CN110195094A (zh) * 2019-05-30 2019-09-03 湖南自兴智慧医疗科技有限公司 一种远程染色体核型检测方法和系统
US20230081581A1 (en) * 2020-01-30 2023-03-16 Termo Health Tecnologia Ltda Mobile system and auxiliary method for evaluating thermographic breast images
CN111310841B (zh) * 2020-02-24 2023-06-20 中南大学湘雅医院 医学图像分类方法、装置、设备、计算机设备和存储介质
CN111310841A (zh) * 2020-02-24 2020-06-19 中南大学湘雅医院 医学图像分类方法、装置、设备、计算机设备和存储介质
CN111275703A (zh) * 2020-02-27 2020-06-12 腾讯科技（深圳）有限公司 图像检测方法、装置、计算机设备和存储介质
CN111275703B (zh) * 2020-02-27 2023-10-27 腾讯科技（深圳）有限公司 图像检测方法、装置、计算机设备和存储介质
US11908154B2 (en) 2021-02-04 2024-02-20 Fibonacci Phyllotaxis Inc. System and method for evaluating tumor stability
USD1041654S1 (en) 2021-02-04 2024-09-10 Fibonacci Phyllotaxis Inc. Transparent tumor size measuring ruler
USD1036665S1 (en) 2021-02-04 2024-07-23 Fibonacci Phyllotaxis Inc. Transparent tumor size measuring ruler with transparent slider
CN114451870A (zh) * 2022-04-12 2022-05-10 中南大学湘雅医院 色素痣恶变风险监测系统
CN115082691A (zh) * 2022-07-28 2022-09-20 吾征智能技术(北京)有限公司 基于深度学习的黑色素瘤识别方法及装置
WO2024104704A1 (fr) * 2022-11-15 2024-05-23 Biotronik Ag Système fondé sur l'ia pour la surveillance de patients souffrant d'une maladie vasculaire périphérique
CN116705325B (zh) * 2023-06-26 2024-01-19 国家康复辅具研究中心 一种伤口感染风险评估方法及其系统
CN116705325A (zh) * 2023-06-26 2023-09-05 国家康复辅具研究中心 一种伤口感染风险评估方法及其系统
GB2632164A (en) * 2023-07-27 2025-01-29 Boots Co Plc Radiance measurement method and system

Also Published As

Publication number Publication date
WO2011087807A3 (fr) 2011-11-17

Similar Documents

Publication Publication Date Title
US8543519B2 (en) System and method for remote melanoma screening
WO2011087807A2 (fr) Système et procédé de dépistage de mélanome à distance
Zhu et al. A deep learning based framework for diagnosing multiple skin diseases in a clinical environment
Guimarães et al. Artificial intelligence in multiphoton tomography: atopic dermatitis diagnosis
Musthafa et al. Optimizing double-layered convolutional neural networks for efficient lung cancer classification through hyperparameter optimization and advanced image pre-processing techniques
Vachmanus et al. DeepMetaForge: A deep vision-transformer metadata-fusion network for automatic skin lesion classification
Baldini et al. An automated approach for real-time informative frames classification in laryngeal endoscopy using deep learning
Aloupogianni et al. Hyperspectral imaging for tumor segmentation on pigmented skin lesions
Zhu et al. A deep learning fusion network trained with clinical and high-frequency ultrasound images in the multi-classification of skin diseases in comparison with dermatologists: a prospective and multicenter study
Youneszade et al. A predictive model to detect cervical diseases using convolutional neural network algorithms and digital colposcopy images
Attallah Skin cancer classification leveraging multi-directional compact convolutional neural network ensembles and gabor wavelets
Arjun et al. Optimizing time prediction and error classification in early melanoma detection using a hybrid RCNN‐LSTM model
Wang et al. Self-supervised learning mechanism for identification of eyelid malignant melanoma in pathologic slides with limited annotation
Mishra et al. A hybrid fused-KNN based intelligent model to access melanoma disease risk using indoor positioning system
Srivastava et al. Binary Classification of Laryngeal Images Utilising ResNet-50 CNN Architecture
Manni et al. Automated tumor assessment of squamous cell carcinoma on tongue cancer patients with hyperspectral imaging
CN120746947A (zh) 一种基于图像数据的肺结核智能识别方法及系统
Parkavi et al. Deep learning techniques for the detection and classification of oral cancer using histopathological images
Kavitha et al. A perspective study on scalable computation model for skin cancer detection: Advancements and challenges
Nazir et al. Integrating artificial intelligence and holographic imaging for advanced cervical cancer diagnosis
Khan et al. Early pigment spot segmentation and classification from iris cellular image analysis with explainable deep learning and multiclass support vector machine
US20250054134A1 (en) Systems and methods for detecting ocular lesions in fundus images
Mir et al. Thyroid Cancer Prediction Using Ensembling Techniques and Multi Modal Data
Benil et al. Detect pre-cancerous tongue lesions for early oral cancer diagnosis using deep learning algorithm
Dadzie et al. Deep learning for automated diagnosis of uveal melanoma

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 10843554

Country of ref document: EP

Kind code of ref document: A2