WO2019200535A1 - 基于人工智能的眼科疾病诊断建模方法、装置及*** - Google Patents

基于人工智能的眼科疾病诊断建模方法、装置及*** Download PDF

Info

Publication number
WO2019200535A1
WO2019200535A1 PCT/CN2018/083393 CN2018083393W WO2019200535A1 WO 2019200535 A1 WO2019200535 A1 WO 2019200535A1 CN 2018083393 W CN2018083393 W CN 2018083393W WO 2019200535 A1 WO2019200535 A1 WO 2019200535A1
Authority
WO
WIPO (PCT)
Prior art keywords
ophthalmic
data set
model
image
artificial intelligence
Prior art date
Application number
PCT/CN2018/083393
Other languages
English (en)
French (fr)
Inventor
刘小青
洪佳旭
倪勇
李双双
王丽丽
何微
郭又文
刘宇轩
刘勇
王威
许睿琦
程静怡
田丽佳
陈文彬
徐讯
Original Assignee
深圳华大生命科学研究院
复旦大学附属眼耳鼻喉科医院
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 深圳华大生命科学研究院, 复旦大学附属眼耳鼻喉科医院 filed Critical 深圳华大生命科学研究院
Priority to EP18914959.4A priority Critical patent/EP3783533A1/en
Priority to PCT/CN2018/083393 priority patent/WO2019200535A1/zh
Priority to US16/967,087 priority patent/US11636340B2/en
Priority to CN201880085521.8A priority patent/CN111656357B/zh
Publication of WO2019200535A1 publication Critical patent/WO2019200535A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/14—Arrangements specially adapted for eye photography
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G06F18/2178—Validation; Performance evaluation; Active pattern learning techniques based on feedback of a supervisor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G06V20/698—Matching; Classification
    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/20—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires
    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • the present application relates to the field of medical technology, and in particular, to an artificial intelligence-based ophthalmic disease diagnosis modeling method, apparatus and system.
  • the artificial intelligence-based ophthalmic disease diagnosis modeling based on related technologies is a differential diagnosis for single diseases, and there is still a big gap between the actual situation of disease diagnosis modeling in which patients and physicians face numerous possibilities in clinical practice.
  • the diagnosis of ophthalmic diseases is not good.
  • the present application aims to solve at least one of the technical problems in the related art to some extent.
  • an object of the present application is to provide an artificial intelligence-based diagnostic method for ophthalmic diseases, which can integrate clinical, ophthalmic imaging, and patient personal information to assist in ophthalmic diagnosis, and can enable artificial intelligence technology to better assist ophthalmology.
  • Disease diagnosis modeling effectively improve the intelligence and accuracy of ophthalmic disease diagnosis modeling, and improve the diagnostic effect.
  • Another object of the present application is to provide an artificial intelligence based ophthalmic disease diagnosis modeling device.
  • Another object of the present application is to propose an artificial intelligence based ophthalmic disease diagnosis modeling system.
  • Another object of the present application is to propose a non-transitory computer readable storage medium.
  • Another object of the present application is to propose an artificial intelligence based ophthalmic disease diagnosis model.
  • Another object of the present application is to provide an artificial intelligence based ophthalmic disease diagnostic test method.
  • Another object of the present application is to propose a computer program product.
  • Another object of the present application is to propose an electronic device.
  • an artificial intelligence-based ophthalmic disease diagnosis modeling method includes: establishing an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set; and training the ophthalmic image data set a first neural network model, obtaining a first classification model; training a second classification model by using the ophthalmologic non-image disease diagnosis questionnaire data set; and combining the first classification model and the second classification model to obtain a target classification network model,
  • the test result output based on the target classification network model is used as a diagnosis result for diagnosis of an ophthalmic disease.
  • the artificial intelligence-based ophthalmic disease diagnosis modeling method proposed by the first aspect of the present application configures and separately trains two sets of data sets, that is, an ophthalmic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • the classification model based on the data of different characteristics, respectively trains different models, and fuses the first classification model and the second classification model respectively trained, and can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmological diagnosis.
  • the artificial intelligence technology can better assist in the diagnosis and diagnosis of ophthalmic diseases, effectively improve the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improve the diagnosis effect.
  • an artificial intelligence-based ophthalmic disease diagnosis modeling apparatus includes: a processor; a memory; storing executable program code in the memory; and the processor is configured to read the memory stored in the memory.
  • the model obtains a target classification network model, and the test result output based on the target classification network model is used as a diagnosis result for diagnosis of an ophthalmic disease.
  • the artificial intelligence-based ophthalmic disease diagnosis modeling device configures and trains two classifications by constructing a multi-dimensional data set, that is, an ophthalmic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • the model trains different models based on data of different characteristics, and fuses the first classification model and the second classification model respectively trained, and can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmological diagnosis, which can make Artificial intelligence technology can better assist in the diagnosis and diagnosis of ophthalmic diseases, effectively improve the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improve the diagnostic effect.
  • an artificial intelligence-based ophthalmic disease diagnosis modeling system includes the artificial intelligence-based ophthalmic disease diagnosis modeling device according to the second aspect.
  • the artificial intelligence-based ophthalmic disease diagnosis modeling system proposed by the third aspect of the present application configures and trains two classifications by constructing a multi-dimensional data set, that is, an ophthalmic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • the model trains different models based on data of different characteristics, and fuses the first classification model and the second classification model respectively trained, and can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmological diagnosis, which can make Artificial intelligence technology can better assist in the diagnosis and diagnosis of ophthalmic diseases, effectively improve the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improve the diagnostic effect.
  • the fourth aspect of the present application provides a non-transitory computer readable storage medium, wherein the storage medium is used to store an application, and the application is used to execute the first application at runtime.
  • the non-transitory computer readable storage medium proposed by the embodiment of the fourth aspect of the present application configures and trains two classification models by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • Different types of data are used to train different models, and the first classification model and the second classification model respectively trained are integrated, which can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmic diagnosis, enabling artificial intelligence.
  • the technology better assists in the diagnosis and diagnosis of ophthalmic diseases, effectively improves the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improves the diagnostic effect.
  • the fifth aspect of the present application provides an artificial intelligence-based ophthalmic disease diagnosis model for performing an artificial intelligence-based ophthalmic disease diagnosis modeling method according to the first aspect of the present application.
  • the artificial intelligence-based ophthalmic disease diagnosis model proposed by the fifth aspect of the present application configures and trains two classification models by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set. Different types of data are used to train different models, and the first classification model and the second classification model respectively trained are integrated, which can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmic diagnosis, enabling artificial intelligence.
  • the technology better assists in the diagnosis and diagnosis of ophthalmic diseases, effectively improves the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improves the diagnostic effect.
  • the sixth aspect of the present application provides an artificial intelligence-based ophthalmic disease diagnosis test method, which uses the artificial intelligence-based ophthalmic disease diagnosis model proposed by the fifth aspect of the present application for diagnostic test. .
  • the artificial intelligence-based ophthalmic disease diagnosis test method proposed in the sixth aspect of the present application configures and trains two classification models by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • Different types of data are used to train different models, and the first classification model and the second classification model respectively trained are integrated, which can integrate clinical and ophthalmic images, and patient personal information to assist in ophthalmic diagnosis, enabling artificial Intelligent technology can better assist in the diagnosis and diagnosis of ophthalmic diseases, effectively improve the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology, and improve the diagnostic effect.
  • the seventh aspect of the present application provides a computer program product, when an instruction in the computer program product is executed by a processor, performing an artificial intelligence based ophthalmic disease diagnosis modeling method.
  • the method includes: establishing an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set; training the first neural network model by using the ophthalmic image data set to obtain a first classification model; and adopting the ophthalmology non-image disease diagnosis questionnaire data set Training a second classification model; combining the first classification model and the second classification model to obtain a target classification network model, and using the test result outputted by the target classification network model as a diagnosis result for diagnosis of an ophthalmic disease .
  • the computer program product proposed in the embodiment of the seventh aspect of the present application configures and trains two classification models based on data sets of different characteristics by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • Different models are trained separately, and the first classification model and the second classification model respectively trained are integrated, which can integrate clinical and ophthalmic imaging, and patient personal information to assist in ophthalmological diagnosis, which can make artificial intelligence technology better assist Diagnostic modeling of ophthalmic diseases, effectively improving the intelligence and accuracy of ophthalmic disease diagnosis modeling, and improving the diagnostic effect.
  • an eighth aspect of the present application provides an electronic device including a housing, a processor, a memory, a circuit board, and a power supply circuit, wherein the circuit board is disposed in a space enclosed by the housing Internally, the processor and the memory are disposed on the circuit board; the power supply circuit is configured to supply power to each circuit or device of the electronic device; the memory is configured to store executable program code; The processor runs a program corresponding to the executable program code by reading executable program code stored in the memory for execution: establishing an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set; Performing the first neural network model on the ophthalmic image data set to obtain a first classification model; training the second classification model by using the ophthalmologic non-image disease diagnosis questionnaire data set; and fusing the first classification model and the second classification model, Obtain a target classification network model and use the test results output by the target classification network model as a diagnosis for ophthalmic diseases To diagnosis.
  • the electronic device proposed in the embodiment of the eighth aspect of the present application configures and trains two classification models by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set, respectively, based on data of different characteristics.
  • Training different models, and combining the first classification model and the second classification model respectively trained can integrate clinical and ophthalmic imaging, and patient personal information to assist in ophthalmological diagnosis, enabling artificial intelligence technology to better assist ophthalmology Disease diagnosis modeling, effectively improve the intelligence and accuracy of ophthalmic disease diagnosis modeling, and improve the diagnostic effect.
  • FIG. 1 is a schematic flow chart of an artificial intelligence-based ophthalmic disease diagnosis modeling method according to an embodiment of the present application
  • 2a is a schematic diagram of initial ophthalmic image data in an embodiment of the present application.
  • 2b is a schematic diagram of the enhanced ophthalmic image data in the embodiment of the present application.
  • 3a is a schematic view showing the characteristics of a macula in the embodiment of the present application.
  • Figure 3b is a schematic view showing the characteristics of the fundus of the fundus in the embodiment of the present application.
  • FIG. 4 is a schematic diagram of a generated confrontation network model
  • FIG. 5 is a schematic flow chart of an artificial intelligence based ophthalmic disease diagnosis modeling method according to another embodiment of the present application.
  • FIG. 6 is a schematic diagram of a fine adjustment process in an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of an artificial intelligence-based ophthalmic disease diagnosis modeling device according to an embodiment of the present application.
  • the artificial intelligence-based ophthalmic disease diagnosis modeling based on related technologies is a differential diagnosis for single diseases, and there is still a big gap between the actual situation of disease diagnosis modeling in which patients and physicians face numerous possibilities in clinical practice.
  • the diagnosis of ophthalmic diseases is not good.
  • the embodiment of the present application proposes an artificial intelligence based ophthalmic disease diagnosis modeling method, which constructs a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set, wherein the ophthalmic non-image disease
  • the diagnostic questionnaire data set is comprehensively considering the symptom information of the ophthalmic disease of the patient and the personal information of the patient, and the embodiment of the present application configures and trains two classification models, and respectively trains different models based on data of different characteristics, and respectively
  • the first classification model and the second classification model obtained by training can be integrated, which can integrate clinical and ophthalmic imaging, and patient personal information to assist in ophthalmological diagnosis, which can make artificial intelligence technology better assist in ophthalmic disease diagnosis modeling and effectively improve ophthalmology.
  • the intelligence and accuracy of all types of disease diagnosis modeling improve diagnostic results.
  • FIG. 1 is a schematic flow chart of an artificial intelligence based ophthalmic disease diagnosis modeling method according to an embodiment of the present application.
  • the method includes:
  • S101 Establish an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set.
  • the ophthalmologic image data set is a data set related to the image collected by the patient's eyes, and may include, for example, clinical data of the patient's eye image and a public data set related to the eye image, which is not limited thereto.
  • the ophthalmic image data set may include: a fundus grayscale image, and an annotation of the target feature in the fundus grayscale image, and the target feature is, for example, a fundus vascular feature or a macula feature, which is not limited thereto.
  • the ophthalmology non-image disease diagnosis questionnaire data set is a reference value for the diagnosis of ophthalmic disease diagnosis, which is selected by the physician based on the ophthalmic diagnosis experience, and includes information on the ophthalmic disease symptoms of the patient and Personal information of the patient.
  • the ophthalmic disease symptom information includes at least one of the following: eye red information, dry eye information, eye pain information, itchy information, foreign body sensation information, burning sensation information, photophobia information, and tearing information, etc. Restriction;
  • the patient's personal information includes at least one of the following: age, gender, duration of illness, and risk factors, etc., without limitation.
  • the embodiments of the present application comprehensively consider the above-mentioned common ophthalmic disease symptom information and patient personal information, and can reduce the model computing resource consumption while improving the accuracy of the ophthalmic disease diagnosis modeling, and improve from another dimension. Diagnostic efficiency. Furthermore, the embodiment of the present application can also be extended on the basis of the above information, and the method is flexible, and the dimension of the model parameter is met to meet the actual clinical personalized diagnosis requirement.
  • the symptom information of the ophthalmic disease can be various symptoms in the actual clinical diagnosis
  • the personal information of the patient can also be the condition of various patients in the actual clinical diagnosis, and thus is not limited to the above, for example,
  • the ophthalmic disease symptom information may also be amblyopia information or far vision information
  • the patient's personal information may also be, for example, drug allergy information.
  • the age of the patient's personal information can be divided into four stages: juvenile (0 to 18 years old), young (19 to 40 years old), middle-aged (40 to 60 years old), and old (over 60 years old);
  • the course of the disease may include, for example, acute, chronic, persistent, or repetitive;
  • high risk factors include, for example, hypertension, diabetes, ocular trauma, familial inheritance, viral infection, immune disease, etc., combined with ophthalmic disease symptom information, in total, there are 37-dimensional factors.
  • the image preprocessing algorithm may be used to enhance the initial ophthalmic image data, and the image preprocessing algorithm is at least one of the following: an algorithm for randomly deforming the image, and cutting the image.
  • the initial ophthalmic image data may be a fundus image of the patient.
  • An example of an image preprocessing algorithm is as follows:
  • Rotation/reflection Rotates an image randomly within a certain angle or changes the orientation of the image content.
  • Zoom Zooms in or out of the image at a certain scale.
  • Shift Shifts an image in a certain way on the image plane. You can specify the panning range and pan step size in a random or artificially defined way, and translate in the horizontal or vertical direction to change the position of the image content.
  • Scale Scales or reduces the image according to the specified scale factor, or refers to the SIFT feature extraction idea, uses the specified scale factor to filter the scale space of the image, and changes the size or blur of the image content.
  • Contrast contrast In the HSV color space of the image, the saturation S and V luminance components are changed, the hue H is kept constant, and the S and V components of each pixel are exponentially operated (exponential factor between 0.25 and 4) , increase the light changes.
  • Noise disturbance Randomly perturbs each pixel of the image, RGB.
  • Commonly used noise modes are salt and pepper noise and Gaussian noise.
  • FIG. 2a is a schematic diagram of initial ophthalmic image data in the embodiment of the present application
  • FIG. 2b is a schematic diagram of enhanced ophthalmic image data in the embodiment of the present application.
  • ophthalmic image after rotation e.g., ophthalmic image after grayscale change
  • ophthalmic image after intensity shearing e.g., ophthalmic image after horizontal flipping
  • ophthalmic image after scale scaling ophthalmic image after horizontal offset
  • adding salt and pepper noise ophthalmic image, the ophthalmic image after vertical flipping, the ophthalmic image after vertical shift, the ophthalmic image after adding Gaussian noise, the ophthalmic image after linear enhancement, and the ophthalmic image after adding Poisson noise.
  • the target feature in the initial ophthalmic image data may be segmented to obtain enhanced image data, and the target feature is a fundus vascular feature or a macula feature.
  • feature enhancement can be performed by means of target detection plus image semantic segmentation.
  • the U-net network is an encoder-decoder structure.
  • the encoder gradually reduces the spatial dimension of the pooling layer, and the decoder gradually repairs the details and spatial dimensions of the object. There is usually a quick connection between the encoder and the decoder, which helps the decoder better fix the details of the target.
  • FIG. 3a is a schematic diagram of the characteristics of the macula in the embodiment of the present application
  • FIG. 3b is a schematic diagram of the characteristics of the fundus of the fundus in the embodiment of the present application, which can be implemented in the ophthalmic image data by segmenting the target features in the initial ophthalmic image data.
  • the characteristics are enhanced to make the characteristic parameters of the ophthalmic image data set more obvious, and improve the accuracy of the diagnosis of ophthalmic diseases.
  • Figure 3a is divided into upper, middle and lower three rows, the first behavior of the fundus grayscale image, the second behavior is the annotation of the target feature (the macular feature) in the fundus grayscale image, and the third behavior is enhanced after the macula feature;
  • Figure 3b is divided into the upper middle
  • the first behavior is the fundus grayscale image
  • the second behavior is the annotation of the target feature (the fundus vascular feature) in the fundus grayscale image
  • the third behavior is enhanced by the fundus vascular feature.
  • the image simulation data corresponding to the generated data set generated by the generated anti-network is also used as the ophthalmologic image data set.
  • the image data corresponding to the enhanced data set may be generated by using a Deep Convolutional Generative Adversarial Networks (DCGAN) model.
  • the DCGAN model is a better performing model that combines a convolutional neural network with an adversarial network.
  • FIG. 4 is a schematic diagram of a generated confrontation network model.
  • the model network consists of a discriminant model and a generated model.
  • Z is the noise, that is, the input of G, which can be Gaussian noise, generally uniform noise; after the G, generate fake image_G(z), then G(z) and X as the input of D, and the output of the last D Indicates the possibility that the data is real, and the value range is 0-1.
  • the image simulation data corresponding to the data set generated by the generated anti-network generation and the enhanced data is used as the ophthalmologic image data set, and the input data of the model training can be quantized, and the operation is convenient, and the ophthalmic image data set and the artificial intelligence are obtained.
  • the deep network model is combined to realize artificial intelligence-assisted diagnosis.
  • some data synthesis algorithm may be used before the deep network model is trained, and more ophthalmic image data is generated by using the existing ophthalmic image data, which can effectively enrich Model training data set.
  • the data synthesis algorithm in the related art for example, the Synthetic Minority Oversampling Technique (SMOTE) algorithm, that is, synthesizes a small number of sample technology, generates more ophthalmic image data.
  • SMOTE Synthetic Minority Oversampling Technique
  • the embodiment of the present application may also optimize the imbalance of the ophthalmic image data set, for example, a sampling method may be adopted.
  • the sampling method is divided into Over Sampling and Under Sampling, and can add a slight random perturbation in the newly generated ophthalmic image dataset, and combine the EasyEnsemble sampling method and the BalanceCascade sampling method to the ophthalmic image data set. Training to optimize processing data imbalance.
  • S103 The second classification model is trained by using an ophthalmologic non-image disease diagnosis questionnaire data set.
  • S104 merging the first classification model and the second classification model to obtain a target classification network model, and using the test result outputted by the target classification network model as a diagnosis result for diagnosis of an ophthalmic disease.
  • the embodiment of the present application separately intercepts the partial convolution layer of the first classification model as a feature of the network feature and the second classification model; adopts a densely connected network or batch normalization.
  • the technology blends features and uses the blended feature training to obtain the target classification network model.
  • a multi-dimensional data set that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set
  • two classification models are configured and trained, and different models are separately trained based on data of different characteristics
  • the first classification model and the second classification model respectively trained to integrate the clinical and ophthalmic images, as well as the patient's personal information to assist in ophthalmological diagnosis can enable artificial intelligence technology to better assist in the diagnosis and modeling of ophthalmic diseases, and effectively improve The intelligence and precision of the diagnosis of all kinds of diseases in ophthalmology enhance the diagnosis.
  • FIG. 5 is a schematic flowchart diagram of an artificial intelligence-based ophthalmic disease diagnosis modeling method according to another embodiment of the present application.
  • the method may further include:
  • S501 training a first neural network model by using an ophthalmic image data set, where the first neural network model comprises: at least two convolution layers, an activation function, at least two fully connected layers, and a sigmoid activation function, each convolution layer connection A pooling layer.
  • S502 migrating the existing existing neural network model trained based on the massive public data set to the supervised learning model, and using the migrated supervised learning model as the first classification model.
  • the first classification model may be implemented by using the method steps of S501 or S502, respectively, or the first classification model may be implemented by using the method steps of S501 or S502, which is not limited.
  • the first neural network model may be first trained using an ophthalmic image data set, the first neural network model including: at least two convolution layers, an activation function, at least two fully connected layers, and a sigmoid activation function.
  • Each convolutional layer is connected to a pooling layer; then the existing neural network model trained based on the massive public data set is migrated to the supervised learning model, specifically, the open mass-based public data set can be migrated.
  • the existing neural network model (such as VGG16 model, VGG19 model, Resnet50 model, and InceptionV3 model, etc.) to the supervised learning model, the application uses the migration learning model for migration, and can realize the ophthalmic image data set in sufficient data volume.
  • the results learned in the state are migrated to a model with a small amount of demand data, which can reduce the computational resource consumption of the model while ensuring the accuracy of the model training.
  • the embodiments of the present application provide a specific execution method in the existing neural network model trained to the publicized data set based on the massive public data set, and the following:
  • Example 1 Learning the bottleneck feature of the first neural network model, using the convolutional layer part of the existing neural network model, and then performing model training of the convolutional layer part based on the collected ophthalmic image data set, the resulting convolution
  • the output of the layer part model is used as the bottleneck feature, and the bottleneck feature is recorded.
  • a supervised learning model is independently trained based on the collected ophthalmic image data set.
  • the supervised learning model is, for example, support vector machine SVM, XGBoost, and fully connected nerve. Network, etc.
  • Table 1 shows the accuracy results of the model based on the existing neural network model VGG19 migration learning to the supervised learning model XGBoost.
  • the ophthalmic image dataset used in this process consisted of 2844 frontal photos and 2595 fundus original images.
  • the ophthalmic image dataset corresponds to the category of ophthalmic diseases: 1169 cases of cataract, 1328 cases of cornea, 865 cases of retina, 865 cases of glaucoma, and the remaining For normal no eye disease pictures.
  • 80% of the ophthalmic image dataset was randomly used for model training, and the remaining 20% of the ophthalmic image dataset was used for model evaluation.
  • the accuracy of the model before and after migration is similar to the four common ophthalmic diseases including ophthalmologic diseases such as cataract, eye corner, glaucoma and retina.
  • the accuracy is close to or exceeds 90%, and is greatly reduced after migration.
  • the amount of data increases the efficiency of model prediction.
  • AUC Accuracy Cataract cataract 0.880 0.875 Corn eye disease 0.985 0.929 Glaucoma glaucoma 0.957 0.890 Retina retinal disease 0.955 0.899
  • the AUC in Table 1 indicates the area under the curve (Area Under the Curve), and the classification model with the larger the AUC value, the higher the correct rate, the AUC value is between 0-1, and if the AUC value is 1, then It indicates that its corresponding classification model is the classification model with the best correct rate.
  • the existing neural network model VGG19 may also be fine-tuned according to the scale of the training data set.
  • FIG. 6 is implemented in the present application. A schematic diagram of the fine-tuning process in the example.
  • the convolutional layer (convolution block one to five) parameters before the last convolutional layer can be frozen, and then the fine-tuned neural network model VGG19 is trained based on the ophthalmic image data set.
  • the degree of migrating learning can be effectively improved.
  • Table 2 shows the accuracy results obtained by the migration fine-tuning training of the VGG19 model. As shown in Table 2, the accuracy of the four common eye diseases is close to or exceed 90%.
  • Example 2 The embodiment of the present application can also use the multi-tasking learning mechanism to migrate the existing existing neural network model trained based on the massive public data set to the supervised learning model. Through the use of multi-tasking learning mechanism for migration learning, it can effectively improve the efficiency of ophthalmic disease diagnosis modeling.
  • Table 3 shows the accuracy results after migration based on the multitasking learning mechanism. For the four common eye diseases, the model performed well and the accuracy was close to or exceeded 90%.
  • Example 3 see Table 4 below.
  • This example uses a hybrid dual model scheme.
  • One network model is based on the VGG19 model and intercepted to the final convolutional layer of the VGG19 model.
  • the other network model is the ResNet50 model, which is also intercepted to ResNet50.
  • the final convolutional layer of the model and then combine the features of the above two networks, and use the merged features to train the partial convolution layer to obtain the target classification network model.
  • the model of the dual model mixture has a certain improvement in the AUC or accuracy of the cataract, cornea and glaucoma of the ophthalmic diseases.
  • the method may further include:
  • S503 Obtain classification information of ophthalmic diseases, and obtain various types of ophthalmic diseases.
  • the common ophthalmic diseases can be divided into 17 categories according to a large class, and each major ophthalmic disease is subdivided into several small types of ophthalmic diseases, a total of 127 subcategories.
  • Each small class of ophthalmic diseases has its corresponding information on common eye symptoms, multiple populations, multiple age groups, disease duration, and high risk factors.
  • each of the small class of eye diseases can be described by a 37-dimensional feature.
  • S504 Using a decision tree algorithm to select a target category of ophthalmic diseases from a plurality of categories of ophthalmic diseases.
  • the target category of ophthalmic diseases is an ophthalmic disease that matches the actual situation, the symptom information of the ophthalmic disease of the patient, and the personal information of the patient.
  • the disease A k always appears on a certain factor, but the patient does not have this factor.
  • the elderly cataract always appears in the elderly population, and the patient is young, the possibility of senile cataract is completely Excluding, therefore, it is possible to exclude ophthalmic diseases of the genital cataract category, and the remaining types of ophthalmic diseases are the ophthalmic diseases of the above-mentioned target categories.
  • the patient suffers from spring.
  • the possibility of ophthalmic diseases in the keratoconjunctivitis category is excluded. Therefore, it is possible to exclude the ophthalmic diseases of the spring keratoconjunctitis type, and the remaining types of ophthalmic diseases as the ophthalmic diseases of the above-mentioned target categories.
  • the second classification model is trained by using the target category of ophthalmic diseases, the patient's ophthalmic disease symptom information, and the patient's personal information.
  • the second classification model is used to analyze the patient's ophthalmic disease symptom information and the patient's personal information, based on the matching degree information under each category of ophthalmic diseases.
  • the calculation method of the second classification model can be illustrated as follows:
  • the matching value of patient B and each type of ophthalmic disease A i is calculated, and the matching values are arranged in descending order, and the first five diseases with the largest matching value are taken as the actual ophthalmic disease prediction of the patient B. result.
  • the calculation formula of the matching value can be selected from the Hamming distance:
  • n 37, corresponding to the 37-dimensional characteristics of each type of ophthalmic diseases, the setting criteria of the weight w j are as shown in Table 5 below:
  • Case situation Weight w j Ocular disease A i may have a certain feature F i , the test case has this feature 1.0 Ocular disease A i may have a certain characteristic F i , and the test case does not have this feature. ⁇ 1.0
  • This application trains the second classification model in the above manner and tests 120 patients with ophthalmic diseases.
  • the prediction is considered correct, and the final test result accuracy is 91%.
  • FIG. 7 is a schematic structural diagram of an artificial intelligence-based ophthalmic disease diagnosis modeling device according to an embodiment of the present application.
  • the apparatus 700 includes: a processor 701; a memory 702; a memory 702 storing executable program code; and a processor 701 executing a program corresponding to the executable program code by reading executable program code stored in the memory 702. For execution:
  • the first neural network model is trained by using an ophthalmic image data set to obtain a first classification model
  • the second classification model is trained by using the ophthalmology non-image disease diagnosis questionnaire data set, and the trained second classification model is used as the second classification model;
  • the first classification model and the second classification model are integrated to obtain a target classification network model, and the test result output based on the target classification network model is used as a diagnosis result for diagnosis of an ophthalmic disease.
  • the processor 701 is further configured to:
  • the first neural network model is trained using an ophthalmic image data set, the first neural network model comprising: at least two convolution layers, an activation function, at least two fully connected layers, and a sigmoid activation function, each convolution layer connected to a pool Layer
  • the existing neural network model trained based on the massive public data set is migrated to the supervised learning model, and the migrated supervised learning model is used as the first classification model.
  • the processor 701 is further configured to:
  • the migration of existing existing neural network models based on massive public data sets to supervised learning models includes:
  • the first neural network model obtained by the fine tuning is migrated into the supervised learning model.
  • the processor 701 is further configured to:
  • the multi-tasking learning mechanism is used to migrate the existing existing neural network models trained based on massive public data sets to the supervised learning model.
  • the processor 701 is further configured to:
  • the partial convolution layer of the first classification model is respectively taken as a feature of the network feature and the second classification model;
  • the feature is merged using densely connected networks or batch normalization techniques, and the feature training after fusion is used to obtain the target classification network model.
  • the processor 701 is further configured to:
  • the image processing algorithm is used to enhance the initial ophthalmic image data.
  • the image preprocessing algorithm is at least one of the following: an algorithm for randomly deforming the image, an algorithm for cutting the image, and compensating for the color or brightness of the image. algorithm;
  • the dataset of the ophthalmologic non-image disease diagnosis questionnaire was directly extracted from the consultation.
  • the processor 701 is further configured to:
  • the image simulation data corresponding to the data set obtained by the enhancement against the network generation is used as the ophthalmic image data set.
  • the ophthalmologic non-image disease diagnosis questionnaire data set includes: ophthalmic disease symptom information of the patient and personal information of the patient, and the processor 701 is further configured to:
  • the decision tree algorithm is used to screen out the target category of ophthalmic diseases from various categories of ophthalmic diseases
  • the second classification model is trained using the target category of ophthalmic diseases, the patient's ophthalmic disease symptom information, and the patient's personal information;
  • the second classification model is used to analyze the patient's ophthalmic disease symptom information and the patient's personal information, based on the matching degree information under each category of ophthalmic diseases.
  • the processor 701 is further configured to:
  • the target features in the initial ophthalmic image data are segmented to obtain enhanced image data, and the target features are fundus vascular features or macular features.
  • a multi-dimensional data set that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set
  • two classification models are configured and trained, and different models are separately trained based on data of different characteristics
  • the first classification model and the second classification model respectively trained to integrate the clinical and ophthalmic images, as well as the patient's personal information to assist in ophthalmological diagnosis can enable artificial intelligence technology to better assist in the diagnosis and modeling of ophthalmic diseases, and effectively improve The intelligence and precision of the diagnosis of all kinds of diseases in ophthalmology enhance the diagnosis.
  • the present application also proposes a storage medium.
  • the storage medium is used to store an application for performing an artificial intelligence-based ophthalmic disease diagnosis modeling method according to an embodiment of the present application at runtime, wherein the artificial intelligence-based ophthalmic disease diagnosis modeling method includes :
  • the first neural network model is trained by using an ophthalmic image data set to obtain a first classification model
  • the second classification model is trained using the ophthalmology non-image disease diagnosis questionnaire data set
  • the first classification model and the second classification model are integrated to obtain a target classification network model, and the test result output based on the target classification network model is used as a diagnosis result for diagnosis of an ophthalmic disease.
  • the application program for performing the artificial intelligence based ophthalmic disease diagnosis modeling method and principle and implementation manner of the embodiment is similar to the artificial intelligence based ophthalmic disease diagnosis modeling method of the above embodiment, in order to avoid redundancy, I won't go into details here.
  • the storage medium of the embodiment of the present application configures and trains two classification models by constructing a multi-dimensional data set, that is, an ophthalmologic image data set and an ophthalmologic non-image disease diagnosis questionnaire data set, and respectively training different models based on data of different characteristics. And the fusion of the first classification model and the second classification model respectively trained to integrate clinical and ophthalmic imaging, as well as patient personal information to assist in ophthalmological diagnosis, enabling artificial intelligence technology to better assist in the diagnosis of ophthalmic diseases It can effectively improve the intelligence and accuracy of the diagnosis and diagnosis of all kinds of diseases in ophthalmology and improve the diagnosis effect.
  • portions of the application can be implemented in hardware, software, firmware, or a combination thereof.
  • multiple steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system.
  • a suitable instruction execution system For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or combination of the following techniques well known in the art: having logic gates for implementing logic functions on data signals. Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.
  • each functional unit in each embodiment of the present application may be integrated into one processing module, or each unit may exist physically separately, or two or more units may be integrated into one module.
  • the above integrated modules can be implemented in the form of hardware or in the form of software functional modules.
  • the integrated modules, if implemented in the form of software functional modules and sold or used as stand-alone products, may also be stored in a computer readable storage medium.
  • the above mentioned storage medium may be a read only memory, a magnetic disk or an optical disk or the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Public Health (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Ophthalmology & Optometry (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Pathology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Probability & Statistics with Applications (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Analysis (AREA)

Abstract

一种基于人工智能的眼科疾病诊断建模方法、装置及***，该方法包括建立眼科图像数据集和眼科非图像疾病诊断问卷数据集(S101)；采用眼科图像数据集训练第一神经网络模型，得到第一分类模型(S102)；采用眼科非图像疾病诊断问卷数据集训练第二分类模型(S103)；融合第一分类模型和第二分类模型，得到目标分类网络模型，并将基于目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果(S104)。通过该方法能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。

Description

基于人工智能的眼科疾病诊断建模方法、装置及*** 技术领域
本申请涉及医疗技术领域，尤其涉及一种基于人工智能的眼科疾病诊断建模方法、装置及***。
背景技术
随着人工智能在医疗领域的应用，越来越多的医疗诊断案例采用了基于人工智能的医疗诊断方法。依托图像识别和深度学习技术，人工智能能够较优地解决医疗领域中，人工处理影像大数据时可能存在的问题。其中，眼科疾病诊断建模作为临床专业分科，其研究对象眼睛是人体中除皮肤外唯一的体表器官，60-70％的眼科疾病可通过图像信息得出诊断，人工智能与医学影像的结合，眼科是一个较好的突破口。
相关技术中基于人工智能的眼科疾病诊断建模，均是面向单种疾病的鉴别诊断，与临床实践中患者与医师要面对无数种可能性的疾病诊断建模的实际情况仍有较大出入，眼科疾病的诊断效果不佳。
发明内容
本申请旨在至少在一定程度上解决相关技术中的技术问题之一。
为此，本申请的一个目的在于提出一种基于人工智能的眼科疾病诊断建模方法，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
本申请的另一个目的在于提出一种基于人工智能的眼科疾病诊断建模装置。
本申请的另一个目的在于提出一种基于人工智能的眼科疾病诊断建模***。
本申请的另一个目的在于提出一种非临时性计算机可读存储介质。
本申请的另一个目的在于提出一种基于人工智能的眼科疾病诊断模型。
本申请的另一个目的在于提出一种基于人工智能的眼科疾病诊断测试方法。
本申请的另一个目的在于提出一种计算机程序产品。
本申请的另一个目的在于提出一种电子设备。
为达到上述目的，本申请第一方面实施例提出的基于人工智能的眼科疾病诊断建模方法，包括：建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
本申请第一方面实施例提出的基于人工智能的眼科疾病诊断建模方法，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并分别训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达到上述目的，本申请第二方面实施例提出的基于人工智能的眼科疾病诊断建模装置，包括：处理器；存储器；存储器内存储可执行程序代码；处理器通过读取存储器中存储的可执行程序代码来运行与可执行程序代码对应的程序，以用于执行：建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型，将所训练得到的第二模型作为所述第二分类模型；融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
本申请第二方面实施例提出的基于人工智能的眼科疾病诊断建模装置，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达到上述目的，本申请第三方面实施例提出的基于人工智能的眼科疾病诊断建模***，包括：上述第二方面实施例提出的基于人工智能的眼科疾病诊断建模装置。
本申请第三方面实施例提出的基于人工智能的眼科疾病诊断建模***，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二 分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达上述目的，本申请第四方面实施例提出了一种非临时性计算机可读存储介质，其中，该存储介质用于存储应用程序，所述应用程序用于在运行时执行本申请第一方面实施例所述的基于人工智能的眼科疾病诊断建模方法。
本申请第四方面实施例提出的非临时性计算机可读存储介质，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达上述目的，本申请第五方面实施例提出了一种基于人工智能的眼科疾病诊断模型，用于执行本申请第一方面实施例提出的基于人工智能的眼科疾病诊断建模方法。
本申请第五方面实施例提出的基于人工智能的眼科疾病诊断模型，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达上述目的，本申请第六方面实施例提出了一种基于人工智能的眼科疾病诊断测试方法，所述方法采用本申请第五方面实施例提出的基于人工智能的眼科疾病诊断模型进行诊断测试。
本申请第六方面实施例提出的基于人工智能的眼科疾病诊断测试方法，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达上述目的，本申请第七方面实施例提出了一种计算机程序产品，当所述计算机程序 产品中的指令由处理器执行时，执行一种基于人工智能的眼科疾病诊断建模方法，所述方法包括：建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
本申请第七方面实施例提出的计算机程序产品，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为达上述目的，本申请第八方面实施例提出了一种电子设备，包括壳体、处理器、存储器、电路板和电源电路，其中，所述电路板安置在所述壳体围成的空间内部，所述处理器和所述存储器设置在所述电路板上；所述电源电路，用于为所述电子设备的各个电路或器件供电；所述存储器用于存储可执行程序代码；所述处理器通过读取所述存储器中存储的可执行程序代码来运行与所述可执行程序代码对应的程序，以用于执行：建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
本申请第八方面实施例提出的电子设备，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
本申请附加的方面和优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本申请的实践了解到。
附图说明
本申请上述的和/或附加的方面和优点从下面结合附图对实施例的描述中将变得明显和 容易理解，其中：
图1是本申请一实施例提出的基于人工智能的眼科疾病诊断建模方法的流程示意图；
图2a为本申请实施例中初始的眼科图像数据示意图；
图2b为本申请实施例中增强后的眼科图像数据示意图；
图3a为本申请实施例中黄斑特征示意图；
图3b为本申请实施例中眼底血管特征示意图；
图4为生成式对抗网络模型示意图；
图5是本申请另一实施例提出的基于人工智能的眼科疾病诊断建模方法的流程示意图；
图6为本申请实施例中微调过程示意图；
图7是本申请一实施例提出的基于人工智能的眼科疾病诊断建模装置的结构示意图。
具体实施方式
下面详细描述本申请的实施例，所述实施例的示例在附图中示出，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的，仅用于解释本申请，而不能理解为对本申请的限制。相反，本申请的实施例包括落入所附加权利要求书的精神和内涵范围内的所有变化、修改和等同物。
相关技术中，随着人工智能在医疗领域的应用，越来越多的医疗诊断案例采用了基于人工智能的医疗诊断方法。依托图像识别和深度学习技术，人工智能能够较优地解决医疗领域中，人工处理影像大数据时可能存在的问题。其中，眼科疾病诊断建模作为临床专业分科，其研究对象眼睛是人体中除皮肤外唯一的体表器官，60-70％的眼科疾病可通过图像信息得出诊断，人工智能与医学影像的结合，眼科是一个较好的突破口。相关技术中基于人工智能的眼科疾病诊断建模，均是面向单种疾病的鉴别诊断，与临床实践中患者与医师要面对无数种可能性的疾病诊断建模的实际情况仍有较大出入，眼科疾病的诊断效果不佳。
因此，本申请实施例提出一种基于人工智能的眼科疾病诊断建模方法，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，其中的眼科非图像疾病诊断问卷数据集为综合考虑病患的眼科疾病症状信息和病患的个人信息，并且，本申请实施例通过配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
图1是本申请一实施例提出的基于人工智能的眼科疾病诊断建模方法的流程示意图。
参见图1，该方法包括：
S101：建立眼科图像数据集和眼科非图像疾病诊断问卷数据集。
其中，眼科图像数据集为对病患眼睛所采集的影像相关的数据集，例如可以包括病患眼睛影像的临床数据和与眼睛影像相关的公开数据集，对此不作限制。
具体地，眼科图像数据集可以包括：眼底灰度图像，以及眼底灰度图像中目标特征的标注，目标特征例如为眼底血管特征或者黄斑特征，对此不作限制。
眼科非图像疾病诊断问卷数据集为医师依据眼科诊断经验，从对病患的问诊信息中筛选出的，对眼科疾病诊断建模有参考价值的信息，例如包括病患的眼科疾病症状信息和病患的个人信息。
可选地，眼科疾病症状信息包括以下至少之一：眼红信息、眼干信息、眼痛信息、眼痒信息、异物感信息、灼烧感信息、畏光信息，以及流泪信息等，对此不作限制；病患的个人信息包括以下至少之一：年龄、性别、病程，以及高危因素等，对此不作限制。
本申请实施例在训练模型过程中，综合考量了上述常见的眼科疾病症状信息和病患个人信息，能够在保障眼科疾病诊断建模精准度的同时，降低模型计算资源消耗，从另一个维度提升诊断效率。进而，本申请实施例也可以在上述信息的基础上进行扩展，方法灵活，实现模型参数的维度满足实际临床的个性化诊断需求。
可以理解的是，由于眼科疾病症状信息可以为实际临床诊断情况中的各种症状，病患的个人信息也可以为实际临床诊断情况中的各种病患的情况，因此不限于上述，例如，眼科疾病症状信息也可以为弱视信息、远视信息，病患的个人信息也可以例如药物过敏信息等。
具体示例如下，病患的个人信息中的年龄可以划分为少年(0～18岁)、青年(19～40岁)、中年(40～60岁)、老年(60岁以上)四个阶段；病程可以例如包括急、慢性、持续，或者反复；高危因素例如包括：高血压、糖尿病、眼外伤、家族遗传、病毒感染、免疫性疾病等，结合眼科疾病症状信息，总计存在37维因素。
本申请实施例在具体执行的过程中，可以采用图像预处理算法对初始的眼科图像数据进行增强，图像预处理算法为以下至少之一：对图像进行随机变形的算法、对图像进行剪切的算法，以及对图像的颜色或者亮度进行补偿的算法，而后，生成与增强得到的图像数据对应的图像模拟数据作为眼科图像数据集，直接从问诊情况中提取眼科非图像疾病诊断问卷数据集。
其中初始的眼科图像数据可以为病患的眼底图像。
图像预处理算法示例如下：
旋转/反射变换(Rotation/reflection)：一定角度内随机旋转图像或改变图像内容的朝向。
翻转变换(flip)：沿着水平或者垂直方向翻转图像。
缩放变换(zoom)：按照一定的比例放大或者缩小图像。
平移变换(shift)：在图像平面上对图像以一定方式进行平移，可以采用随机或人为定义的方式指定平移范围和平移步长，沿水平或竖直方向进行平移，改变图像内容的位置。
尺度变换(scale)：对图像按照指定的尺度因子，进行放大或缩小，或者参照SIFT特征提取思想，利用指定的尺度因子对图像滤波构造尺度空间，改变图像内容的大小或模糊程度。
对比度变换(contrast)：在图像的HSV颜色空间，改变饱和度S和V亮度分量，保持色调H不变，对每个像素的S和V分量进行指数运算(指数因子在0.25到4之间)，增加光照变化。
噪声扰动(noise)：对图像的每个像素RGB进行随机扰动，常用的噪声模式是椒盐噪声和高斯噪声。
参见图2，图2a为本申请实施例中初始的眼科图像数据示意图,图2b为本申请实施例中增强后的眼科图像数据示意图。包括：旋转后的眼科图像、灰度改变后的眼科图像、强度剪切后的眼科图像、水平翻转后的眼科图像、尺度缩放后的眼科图像、水平偏移后的眼科图像、添加椒盐噪声后的眼科图像、垂直翻转后的眼科图像、垂直偏移后的眼科图像、添加高斯噪声后的眼科图像、线性增强后的眼科图像、添加泊松噪声后的眼科图像。
可选地，本申请实施例在具体执行的过程中，还可以对初始的眼科图像数据中的目标特征进行分割，以得到增强得到的图像数据，目标特征为眼底血管特征或者黄斑特征。
例如，可以采用目标检测加上图像语义分割的手段来进行特征增强。以U-net网络对眼睛图像做语义分割为例，U-net网络是编码器-解码器结构。编码器逐渐减少池化层的空间维度，解码器逐步修复物体的细节和空间维度。编码器和解码器之间通常存在快捷连接，因此能帮助解码器更好地修复目标的细节。
参见图3a，图3a为本申请实施例中黄斑特征示意图，图3b为本申请实施例中眼底血管特征示意图，通过对初始的眼科图像数据中的目标特征进行分割，能够实现对眼科图像数据中的特征进行增强，使得眼科图像数据集中的特征参数更加明显，提升眼科疾病诊断建模精准度。图3a分为上中下三行，第一行为眼底灰度图像，第二行为对眼底灰度图像中目标特征(黄斑特征)的标注，第三行为增强后黄斑特征；图3b分为上中下三行，第一行为眼底 灰度图像，第二行为对眼底灰度图像中目标特征(眼底血管特征)的标注，第三行为增强后眼底血管特征。
本申请实施例在具体执行的过程中，还可以采用生成式对抗网络生成与增强得到的数据集对应的图像模拟数据作为眼科图像数据集。
作为一种示意，本申请实施例中可以采用生成式对抗网络(Deep Convolutional Generative Adversarial Networks，DCGAN)模型生成与增强得到的数据集对应的图像模拟数据。DCGAN模型是将卷积神经网络和对抗网络结合起来一个表现较好的模型。
参见图4，图4为生成式对抗网络模型示意图。该模型网络是由一个判别模型和生成模型组成。Z是噪声，也就是G的输入，可以是高斯噪声，一般为均匀噪声；经过G之后生成fake image—G(z),然后将G(z)和X作为D的输入，最后的D的输出表示该数据为real的可能性，该值范围0-1。
本申请实施例中通过采用生成式对抗网络生成与增强得到的数据集对应的图像模拟数据作为眼科图像数据集，能够对模型训练的输入数据进行量化，便于运算，将眼科图像数据集和人工智能中的深度网络模型进行结合，实现人工智能辅助诊断。
另外，本申请实施例中为了保障眼科图像数据集的参考价值，还可以在训练深度网络模型之前，采用了一些数据合成法算法，利用已有眼科图像数据生成更多眼科图像数据，能够有效丰富模型训练的数据集。
本申请实施例在具体执行的过程中，可以采用相关技术中的数据合成算法，例如，(Synthetic Minority Oversampling Technique，SMOTE)算法，即合成少数样本类技术，生成更多眼科图像数据。
本申请实施例在具体执行的过程中，还可以对眼科图像数据集的不平衡进行优化处理，例如，可以采取采样法。采样法分为过采样(Over Sampling)和欠采样(Under Sampling)，并可以在新生成的眼科图像数据集中加入轻微的随机扰动，以及，结合EasyEnsemble采样法和BalanceCascade采样法对眼科图像数据集进行训练以优化处理数据不平衡。
S102：采用眼科图像数据集训练第一神经网络模型，得到第一分类模型。
S103：采用眼科非图像疾病诊断问卷数据集训练第二分类模型。
S102和S103的执行过程可以详见下述实施例。
S104：融合第一分类模型和第二分类模型，得到目标分类网络模型，并将基于目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
可选地，一些实施例中，本申请实施例在具体执行的过程中，分别截取第一分类模型的 部分卷积层为网络特征和第二分类模型的特征；采用密集连接网络或者批量归一化技术融合特征，并采用融合后的特征训练，以得到目标分类网络模型。
本实施例中，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
可选地，一些实施例中，作为另一种示例：
参见图5，图5是本申请另一实施例提出的基于人工智能的眼科疾病诊断建模方法的流程示意图。针对上述实施例中步骤S102，还可以具体包括：
S501：采用眼科图像数据集训练第一神经网络模型，第一神经网络模型包括：至少两个的卷积层、激活函数、至少两个全连接层，以及sigmoid激活函数，每个卷积层连接一个池化层。
S502：迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，并将迁移后的有监督学习模型作为第一分类模型。
本发明实施例中，可以分别采用上述S501或者S502的方法步骤实现得到第一分类模型，或者，也可以采用结合上述S501或者S502的方法步骤实现得到第一分类模型，对此不作限制。
在具体执行的过程中，可以首先采用眼科图像数据集训练第一神经网络模型，第一神经网络模型包括：至少两个的卷积层、激活函数、至少两个全连接层，以及sigmoid激活函数，每个卷积层连接一个池化层；而后迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，具体地，可以迁移公开的基于海量公开数据集所训练的现有神经网络模型(具体如VGG16模型、VGG19模型、Resnet50模型，以及InceptionV3模型等)至有监督学习模型中，本申请利用迁移学习模型进行迁移，可以实现将眼科图像数据集在数据量充足状态下学习到的结果，迁移至需求数据量较小模型中，能够在保证模型训练精准度的同时，降低模型计算资源消耗。
作为一种示例，本申请实施例提供了迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中的具体执行方法，参见下述：
示例一：学习第一神经网络模型的瓶颈特征(bottleneck)，运用现有神经网络模型的卷积层部分，然后基于收集的眼科图像数据集进行卷积层部分的模型训练，将得到的卷积层 部分的模型的输出作为瓶颈特征，并对瓶颈特征进行记录，而后基于收集的眼科图像数据集独立训练一个有监督学习模型，有监督学习模型例如，支持向量机SVM、XGBoost，以及全连接神经网络等。
参见下述表1，表1所示为基于现有神经网络模型VGG19迁移学习至有监督学习模型XGBoost中训练得到的模型的准确度结果。此过程采用的眼科图像数据集为2844张前节照及2595张眼底原始图片，其中，眼科图像数据集对应眼科疾病类别：白内障1169例，角膜病例1328，视网膜病例865，青光眼病例865例，剩余为正常无眼病图片。在训练的过程中，随机采用80％的眼科图像数据集用于模型训练，剩余20％的眼科图像数据集用于模型评估。如表1所示，对包括眼科疾病类别白内障，眼角，青光眼及视网膜在内的四种常见眼科疾病，模型迁移前后的准确度较为接近，准确度均接近或超过90％，且迁移后大幅降低了数据量，提升模型预测效率。
表1
  AUC 准确度
Cataract白内障 0.880 0.875
Corn眼角疾病 0.985 0.929
Glaucoma青光眼 0.957 0.890
Retina视网膜疾病 0.955 0.899
其中，表1中的AUC表示曲线下面积(Area Under the Curve)，AUC值越大的分类模型，其正确率越高，AUC取值在0-1之间，若AUC取值为1，则表示其对应的分类模型为正确率最佳的分类模型。
进一步地，为了进一步提升上述模型的准确度，本申请实施例中，还可以根据训练数据集的规模，也可以适当对现有神经网络模型VGG19进行微调，参见图6，图6为本申请实施例中微调过程示意图。
1)搭建神经网络模型VGG19并载入预设权重。
2)将预设的全连接网络加在神经网络模型VGG19的顶部并载入上述的权重。
3)冻结神经网络模型VGG19的部分参数。
例如，参见图6，可以将最后一个卷积层之前的卷积层(卷积块一至五)参数冻结，而后，基于眼科图像数据集训练微调后的神经网络模型VGG19。通过适当对现有神经网络模型进行微调，能够有效提升迁移学习的充分程度。
基于与表1相同的数据，表2为VGG19模型经过迁移微调训练得到的准确度结果。如表 2所示，对四种常见眼科疾病，准确度均接近或超过90％。
表2
  AUC 准确度
Cataract白内障 0.907 0.882
Corn眼角疾病 0.989 0.946
Glaucoma青光眼 0.938 0.861
Retina视网膜疾病 0.954 0.907
示例二：本申请实施例还可以采用多任务学习机制迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中。通过采用多任务学习机制进行迁移学习，能够有效提升眼科疾病诊断建模的效率。
表3所示为基于多任务学习机制迁移后的准确度结果。针对四种常见眼科疾病，模型表现良好，准确度均接近或超过90％。
表3
  AUC 准确度
Cataract白内障 0.947 0.876
Corn眼角疾病 0.986 0.948
Glaucoma青光眼 0.927 0.885
Retina视网膜疾病 0.917 0.859
示例三，参见下述表4，本实例采用混合双模型方案，其中一个网络模型以VGG19模型为基础，截取至VGG19模型的最后的卷积层，另一个网络模型为ResNet50模型，同样截取至ResNet50模型的最后的卷积层，而后，将上述两个网络的特征进行合并，并采用并合后的特征训练部分卷积层，以得到目标分类网络模型。
如表4所示，相对于单模型方案来说，双模型混合后的模型，对于眼科疾病白内障、角膜、青光眼的AUC或准确度都有一定的提升。
表4
  AUC 准确度
Cataract白内障 0.936 0.882
Corn眼角疾病 0.991 0.958
Glaucoma青光眼 0.938 0.879
Retina视网膜疾病 0.943 0.867
针对上述实施例中步骤S103，还可以具体包括：
S503：获取眼科疾病的分类信息，得到多种类别的眼科疾病。
本申请实施例在执行的过程中，可以将常见的眼科疾病按大类分为17类，将每个大类眼科疾病细分为若干小类眼科疾病，共127个小类。每个小类眼科疾病有其对应的常见眼部症状信息，多发人群、多发年龄段、病程、高危因素。
因此，如上眼科疾病症状信息和病患的个人信息，每个小类眼科疾病可用一个37维的特征来描述。
S504：采用决策树算法从多种类别的眼科疾病中筛选出目标类别的眼科疾病。
其中，目标类别的眼科疾病为结合实际情况，与病患的眼科疾病症状信息，以及病患的个人信息所匹配的眼科疾病。
实际情况例如，疾病A k在某个因素上总是出现，但病患没有出现该因素，例如，老年型白内障总是出现老年人群，而病患为青年，则患老年型白内障可能性被完全排除，因此，可以将排除老年型白内障类别眼科疾病，所剩余类别的眼科疾病作为上述的目标类别的眼科疾病。
又例如，病患中出现某个因素，但疾病A k不会出现该因素的情况，例如，病患出现黑影飘动，但是春季角结膜炎不会出现该症状，则该病患，患春季角结膜炎类别眼科疾病的可能性被排除，因此，可以将排除春季角结膜炎类别眼科疾病，所剩余类别的眼科疾病作为上述的目标类别的眼科疾病。
S505：采用目标类别的眼科疾病、病患的眼科疾病症状信息，以及病患的个人信息训练第二分类模型。
其中，第二分类模型用于分析病患的眼科疾病症状信息和病患的个人信息，基于每种类别的眼科疾病下的匹配度信息。
第二分类模型的计算方法可以示意如下：
对于目标类别的眼科疾病，则采用计算病患B与每类眼科疾病A i的匹配值，对匹配值降序排列，取匹配值最大的前五个疾病，作为该病患B的实际眼科疾病预测结果。匹配值的计算公式可以选用海明距离：
Figure PCTCN2018083393-appb-000001
其中，n＝37，对应每类眼科疾病的37维的特征，权重w j的设定标准如下表5所示：
表5
病例情况 权重w j
眼科疾病A i可能会出现某特征F i，测试病例有该特征的情况 1.0
眼科疾病A i可能会出现某特征F i，测试病例没有该特征的情况 <1.0
本申请采用上述方式训练第二分类模型，并对120种眼科疾病的病患进行测试。以预测的5类眼科疾病中，若包含该病患的真实诊断的眼科疾病，则视为预测正确，最终的测试结果准确率为91％。
图7是本申请一实施例提出的基于人工智能的眼科疾病诊断建模装置的结构示意图。
参见图7，该装置700包括：处理器701；存储器702；存储器702内存储可执行程序代码；处理器701通过读取存储器702中存储的可执行程序代码来运行与可执行程序代码对应的程序，以用于执行：
建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
采用眼科图像数据集训练第一神经网络模型，得到第一分类模型；
采用眼科非图像疾病诊断问卷数据集训练第二分类模型，将所训练得到的第二分类模型作为第二分类模型；
融合第一分类模型和第二分类模型，得到目标分类网络模型，并将基于目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
可选地，一些实施例中，处理器701，还用于：
采用眼科图像数据集训练第一神经网络模型，第一神经网络模型包括：至少两个的卷积层、激活函数、至少两个全连接层，以及sigmoid激活函数，每个卷积层连接一个池化层；
迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，并将迁移后的有监督学习模型作为第一分类模型。
可选地，一些实施例中，处理器701，还用于：
对第一神经网络模型中卷积层的参数进行微调处理，并采用眼科图像数据集训练所微调得到的第一神经网络模型；
迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，包括：
迁移微调得到的第一神经网络模型至有监督学习模型中。
可选地，一些实施例中，处理器701，还用于：
采用多任务学习机制迁移公开的基于海量公开数据集所训练的现有神经网络模型至有 监督学习模型中。
可选地，一些实施例中，处理器701，还用于：
分别截取第一分类模型的部分卷积层为网络特征和第二分类模型的特征；
采用密集连接网络或者批量归一化技术融合特征，并采用融合后的特征训练，以得到目标分类网络模型。
可选地，一些实施例中，处理器701，还用于：
采用图像预处理算法对初始的眼科图像数据进行增强，图像预处理算法为以下至少之一：对图像进行随机变形的算法、对图像进行剪切的算法，以及对图像的颜色或者亮度进行补偿的算法；
生成与增强得到的图像数据对应的图像模拟数据作为眼科图像数据集；
直接从问诊情况中提取眼科非图像疾病诊断问卷数据集。
可选地，一些实施例中，处理器701，还用于：
采用生成式对抗网络生成与增强得到的数据集对应的图像模拟数据作为眼科图像数据集。
可选地，一些实施例中，其特征在于，眼科非图像疾病诊断问卷数据集包括：病患的眼科疾病症状信息和病患的个人信息，处理器701，还用于：
获取眼科疾病的分类信息，得到多种类别的眼科疾病；
采用决策树算法从多种类别的眼科疾病中筛选出目标类别的眼科疾病；
采用目标类别的眼科疾病、病患的眼科疾病症状信息，以及病患的个人信息训练第二分类模型；
其中，第二分类模型用于分析病患的眼科疾病症状信息和病患的个人信息，基于每种类别的眼科疾病下的匹配度信息。
可选地，一些实施例中，处理器701，还用于：
对初始的眼科图像数据中的目标特征进行分割，以得到增强得到的图像数据，目标特征为眼底血管特征或者黄斑特征。
需要说明的是，前述图1-图6实施例中对基于人工智能的眼科疾病诊断建模方法实施例的解释说明也适用于该实施例的基于人工智能的眼科疾病诊断建模装置700，其实现原理类似，此处不再赘述。
本实施例中，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分 别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
为了实现上述实施例，本申请还提出一种存储介质。其中，该存储介质用于存储应用程序，该应用程序用于在运行时执行本申请实施例的基于人工智能的眼科疾病诊断建模方法，其中，该基于人工智能的眼科疾病诊断建模方法包括：
建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
采用眼科图像数据集训练第一神经网络模型，得到第一分类模型；
采用眼科非图像疾病诊断问卷数据集训练第二分类模型；
融合第一分类模型和第二分类模型，得到目标分类网络模型，并将基于目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
需要说明的是，本实施例的应用程序执行基于人工智能的眼科疾病诊断建模方法和原理和实现方式与上述实施例的基于人工智能的眼科疾病诊断建模方法类似，为了避免冗余，此处不再赘述。
本申请实施例的存储介质，通过构建多维度的数据集，即，眼科图像数据集和眼科非图像疾病诊断问卷数据集，配置并训练两个分类模型，基于不同特性的数据分别训练不同的模型，以及将分别训练得到的第一分类模型和第二分类模型进行融合，能够集成临床、眼科影像，以及病患个人信息辅助进行眼科诊断，能够使得人工智能技术更好地辅助眼科疾病诊断建模，有效提升眼科全种类疾病诊断建模的智能化和精准度，提升诊断效果。
需要说明的是，在本申请的描述中，术语“第一”、“第二”等仅用于描述目的，而不能理解为指示或暗示相对重要性。此外，在本申请的描述中，除非另有说明，“多个”的含义是两个或两个以上。
流程图中或在此以其他方式描述的任何过程或方法描述可以被理解为，表示包括一个或更多个用于实现特定逻辑功能或过程的步骤的可执行指令的代码的模块、片段或部分，并且本申请的优选实施方式的范围包括另外的实现，其中可以不按所示出或讨论的顺序，包括根据所涉及的功能按基本同时的方式或按相反的顺序，来执行功能，这应被本申请的实施例所属技术领域的技术人员所理解。
应当理解，本申请的各部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中，多个步骤或方法可以用存储在存储器中且由合适的指令执行***执行的软件或固件来实现。例如，如果用硬件来实现，和在另一实施方式中一样，可用本领域公知的下列技术 中的任一项或他们的组合来实现：具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路，具有合适的组合逻辑门电路的专用集成电路，可编程门阵列(PGA)，现场可编程门阵列(FPGA)等。
本技术领域的普通技术人员可以理解实现上述实施例方法携带的全部或部分步骤是可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，该程序在执行时，包括方法实施例的步骤之一或其组合。
此外，在本申请各个实施例中的各功能单元可以集成在一个处理模块中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时，也可以存储在一个计算机可读取存储介质中。
上述提到的存储介质可以是只读存储器，磁盘或光盘等。
在本说明书的描述中，参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本申请的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不一定指的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任何的一个或多个实施例或示例中以合适的方式结合。
尽管上面已经示出和描述了本申请的实施例，可以理解的是，上述实施例是示例性的，不能理解为对本申请的限制，本领域的普通技术人员在本申请的范围内可以对上述实施例进行变化、修改、替换和变型。

Claims (26)

  1. 一种基于人工智能的眼科疾病诊断建模方法，其特征在于，包括以下步骤：
    建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
    采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；
    采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；
    融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
  2. 如权利要求1所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型，包括：
    采用所述眼科图像数据集训练第一神经网络模型，所述第一神经网络模型包括：至少两个的卷积层、激活函数、至少两个全连接层，以及sigmoid激活函数，所述每个卷积层连接一个池化层；或者，
    迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，并将迁移后的有监督学习模型作为所述第一分类模型。
  3. 如权利要求2所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述采用所述眼科图像数据集训练第一神经网络模型之后，还包括：
    对所述第一神经网络模型中卷积层的参数进行微调处理，并采用所述眼科图像数据集训练所微调得到的第一神经网络模型；
    所述迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，包括：
    迁移所述微调得到的第一神经网络模型至所述有监督学习模型中。
  4. 如权利要求2或3所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，包括：
    采用多任务学习机制迁移公开的基于海量公开数据集所训练的现有神经网络模型至所述有监督学习模型中。
  5. 如权利要求1所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，包括：
    分别截取所述第一分类模型的部分卷积层为网络特征和所述第二分类模型的特征；
    采用密集连接网络或者批量归一化技术融合所述特征，并采用融合后的特征训练，以得 到所述目标分类网络模型。
  6. 如权利要求1所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述建立眼科图像数据集和眼科非图像疾病诊断问卷数据集，包括：
    采用图像预处理算法对初始的眼科图像数据进行增强，所述图像预处理算法为以下至少之一：对图像进行随机变形的算法、对图像进行剪切的算法，以及对图像的颜色或者亮度进行补偿的算法；
    生成与增强得到的图像数据对应的图像模拟数据作为所述眼科图像数据集；
    直接从问诊情况中提取所述眼科非图像疾病诊断问卷数据集。
  7. 如权利要求6所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述生成与增强得到的数据集对应的图像模拟数据作为所述眼科图像数据集，包括：
    采用生成式对抗网络生成与增强得到的数据集对应的图像模拟数据作为所述眼科图像数据集。
  8. 如权利要求1或6所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述眼科非图像疾病诊断问卷数据集包括：病患的眼科疾病症状信息和病患的个人信息，所述采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型，包括：
    获取所述眼科疾病的分类信息，得到多种类别的眼科疾病；
    采用决策树算法从所述多种类别的眼科疾病中筛选出目标类别的眼科疾病；
    采用所述目标类别的眼科疾病、所述病患的眼科疾病症状信息，以及所述病患的个人信息迭代训练所述第二分类模型；
    其中，所述第二分类模型用于分析所述病患的眼科疾病症状信息和所述病患的个人信息，基于每种类别的眼科疾病下的匹配度信息。
  9. 如权利要求6所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述采用图像预处理算法对初始的眼科图像数据进行增强，还包括：
    对所述初始的眼科图像数据中的目标特征进行分割，以得到所述增强得到的图像数据，所述目标特征为眼底血管特征或者黄斑特征。
  10. 如权利要求8所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述眼科疾病症状信息包括以下至少之一：眼红信息、眼干信息、眼痛信息、眼痒信息、异物感信息、灼烧感信息、畏光信息，以及流泪信息。
  11. 如权利要求8所述的基于人工智能的眼科疾病诊断建模方法，其特征在于，所述病患的个人信息包括以下至少之一：年龄、性别、病程，以及高危因素。
  12. 一种基于人工智能的眼科疾病诊断建模装置，其特征在于，包括：
    处理器；
    存储器；
    存储器内存储可执行程序代码；处理器通过读取存储器中存储的可执行程序代码来运行与可执行程序代码对应的程序，以用于执行：
    建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
    采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；
    采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；
    融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
  13. 如权利要求12所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    采用所述眼科图像数据集训练第一神经网络模型，所述第一神经网络模型包括：至少两个的卷积层、激活函数、至少两个全连接层，以及sigmoid激活函数，所述每个卷积层连接一个池化层；或者，
    迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，并将迁移后的有监督学习模型作为所述第一分类模型。
  14. 如权利要求13所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    对所述第一神经网络模型中卷积层的参数进行微调处理，并采用所述眼科图像数据集训练所微调得到的第一神经网络模型；
    所述迁移公开的基于海量公开数据集所训练的现有神经网络模型至有监督学习模型中，包括：
    迁移所述微调得到的第一神经网络模型至所述有监督学习模型中。
  15. 如权利要求13或14所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    采用多任务学习机制迁移公开的基于海量公开数据集所训练的现有神经网络模型至所述有监督学习模型中。
  16. 如权利要求12所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    分别截取所述第一分类模型的部分卷积层为网络特征和所述第二分类模型的特征；
    采用密集连接网络或者批量归一化技术融合所述特征，并采用融合后的特征训练，以得到所述目标分类网络模型。
  17. 如权利要求12所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    采用图像预处理算法对初始的眼科图像数据进行增强，所述图像预处理算法为以下至少之一：对图像进行随机变形的算法、对图像进行剪切的算法，以及对图像的颜色或者亮度进行补偿的算法；
    生成与增强得到的图像数据对应的图像模拟数据作为所述眼科图像数据集；
    直接从问诊情况中提取所述眼科非图像疾病诊断问卷数据集。
  18. 如权利要求17所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    采用生成式对抗网络生成与增强得到的数据集对应的图像模拟数据作为所述眼科图像数据集。
  19. 如权利要求12或17所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述眼科非图像疾病诊断问卷数据集包括：病患的眼科疾病症状信息和病患的个人信息，所述处理器，还用于：
    获取所述眼科疾病的分类信息，得到多种类别的眼科疾病；
    采用决策树算法从所述多种类别的眼科疾病中筛选出目标类别的眼科疾病；
    采用所述目标类别的眼科疾病、所述病患的眼科疾病症状信息，以及所述病患的个人信息迭代训练所述第二分类模型；
    其中，所述第二分类模型用于分析所述病患的眼科疾病症状信息和所述病患的个人信息，基于每种类别的眼科疾病下的匹配度信息。
  20. 如权利要求17所述的基于人工智能的眼科疾病诊断建模装置，其特征在于，所述处理器，还用于：
    对所述初始的眼科图像数据中的目标特征进行分割，以得到所述增强得到的图像数据，所述目标特征为眼底血管特征或者黄斑特征。
  21. 一种基于人工智能的眼科疾病诊断建模***，其特征在于，包括：
    如权利要求12-20任一项所述的基于人工智能的眼科疾病诊断建模装置。
  22. 一种非临时性计算机可读存储介质，具有存储于其中的指令，当所述指令被电子设 备的处理器执行时，所述处理器执行如权利要求1-11任一项所述的基于人工智能的眼科疾病诊断建模方法。
  23. 一种基于人工智能的眼科疾病诊断模型，其特征在于，包括：
    所述模型用于执行如上述权利要求1-11中任一项所述的基于人工智能的眼科疾病诊断建模方法。
  24. 一种基于人工智能的眼科疾病诊断测试方法，其特征在于，其中，
    所述方法采用如上述权利要求23中的基于人工智能的眼科疾病诊断模型进行诊断测试。
  25. 一种计算机程序产品，当所述计算机程序产品中的指令由处理器执行时，执行一种基于人工智能的眼科疾病诊断建模方法，所述方法包括：
    建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
    采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；
    采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；
    融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
  26. 一种电子设备，包括壳体、处理器、存储器、电路板和电源电路，其中，所述电路板安置在所述壳体围成的空间内部，所述处理器和所述存储器设置在所述电路板上；所述电源电路，用于为所述电子设备的各个电路或器件供电；所述存储器用于存储可执行程序代码；所述处理器通过读取所述存储器中存储的可执行程序代码来运行与所述可执行程序代码对应的程序，以用于执行：
    建立眼科图像数据集和眼科非图像疾病诊断问卷数据集；
    采用所述眼科图像数据集训练第一神经网络模型，得到第一分类模型；
    采用所述眼科非图像疾病诊断问卷数据集训练第二分类模型；
    融合所述第一分类模型和所述第二分类模型，得到目标分类网络模型，并将基于所述目标分类网络模型输出的测试结果作为对眼科疾病进行诊断得到的诊断结果。
PCT/CN2018/083393 2018-04-17 2018-04-17 基于人工智能的眼科疾病诊断建模方法、装置及*** WO2019200535A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
EP18914959.4A EP3783533A1 (en) 2018-04-17 2018-04-17 Artificial intelligence-based ophthalmic disease diagnostic modeling method, apparatus, and system
PCT/CN2018/083393 WO2019200535A1 (zh) 2018-04-17 2018-04-17 基于人工智能的眼科疾病诊断建模方法、装置及***
US16/967,087 US11636340B2 (en) 2018-04-17 2018-04-17 Modeling method and apparatus for diagnosing ophthalmic disease based on artificial intelligence, and storage medium
CN201880085521.8A CN111656357B (zh) 2018-04-17 2018-04-17 眼科疾病分类模型的建模方法、装置及***

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/083393 WO2019200535A1 (zh) 2018-04-17 2018-04-17 基于人工智能的眼科疾病诊断建模方法、装置及***

Publications (1)

Publication Number Publication Date
WO2019200535A1 true WO2019200535A1 (zh) 2019-10-24

Family

ID=68239305

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/083393 WO2019200535A1 (zh) 2018-04-17 2018-04-17 基于人工智能的眼科疾病诊断建模方法、装置及***

Country Status (4)

Country Link
US (1) US11636340B2 (zh)
EP (1) EP3783533A1 (zh)
CN (1) CN111656357B (zh)
WO (1) WO2019200535A1 (zh)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110840428A (zh) * 2019-11-29 2020-02-28 苏州大学 基于一维U-Net网络的无创血压估计方法
CN111048210A (zh) * 2019-12-31 2020-04-21 上海鹰瞳医疗科技有限公司 基于眼底图像评估疾病风险的方法及设备
CN111415741A (zh) * 2020-03-05 2020-07-14 北京深睿博联科技有限责任公司 基于隐式表观学习的乳腺x射线影像分类模型训练方法
CN111681756A (zh) * 2020-05-29 2020-09-18 吾征智能技术(北京)有限公司 基于痰液性状认知的疾病征兆预测***
CN112185564A (zh) * 2020-10-20 2021-01-05 福州数据技术研究院有限公司 一种基于结构化电子病历的眼科疾病预测方法和存储设备
CN112784801A (zh) * 2021-02-03 2021-05-11 紫东信息科技(苏州)有限公司 基于文本和图片的双模态胃部疾病分类方法及装置
CN112884729A (zh) * 2021-02-04 2021-06-01 北京邮电大学 基于双模态深度学习的眼底疾病辅助诊断方法和装置
CN113096804A (zh) * 2021-04-08 2021-07-09 武汉爱尔眼科医院有限公司 一种干眼症患者数据统计***
CN113421638A (zh) * 2021-06-22 2021-09-21 平安科技（深圳）有限公司 基于迁移学习的模型生成方法、装置和计算机设备
CN113435607A (zh) * 2021-06-09 2021-09-24 内蒙古大学 一种基于联邦学习的疾病筛查方法
CN113707301A (zh) * 2021-08-30 2021-11-26 康键信息技术（深圳）有限公司 基于人工智能的远程问诊方法、装置、设备及介质
CN115631367A (zh) * 2022-09-30 2023-01-20 中国医学科学院生物医学工程研究所 基于神经网络模型的眼科超声图像分类方法以及装置

Families Citing this family (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019232742A1 (zh) * 2018-06-07 2019-12-12 深圳迈瑞生物医疗电子股份有限公司 基于超声影像设备的图像分析方法及超声影像设备
US10713542B2 (en) * 2018-10-24 2020-07-14 The Climate Corporation Detection of plant diseases with multi-stage, multi-scale deep learning
EP3948768A1 (en) * 2019-03-26 2022-02-09 Panakeia Technologies Limited A method of processing an image of tissue, a system for processing an image of tissue, a method for disease diagnosis and a disease diagnosis system
TWI705458B (zh) * 2019-09-19 2020-09-21 沐恩生醫光電股份有限公司 一種醫療影像辨識的方法與系統
CN111261288A (zh) * 2020-04-07 2020-06-09 上海市精神卫生中心（上海市心理咨询培训中心） 一种基于bdnf早期识别双相障碍的方法
KR20210142930A (ko) * 2020-05-19 2021-11-26 삼성에스디에스 주식회사 퓨 샷 학습 방법 및 이를 수행하기 위한 장치
CN111709371B (zh) * 2020-06-17 2023-12-22 腾讯科技（深圳）有限公司 基于人工智能的分类方法、装置、服务器和存储介质
CN112233087A (zh) * 2020-10-14 2021-01-15 武汉楚精灵医疗科技有限公司 一种基于人工智能的眼科超声疾病诊断方法和***
WO2022139943A2 (en) * 2020-10-23 2022-06-30 Remmie, Inc. Machine learning for ear disease diagnosis assistance
CN112418303B (zh) * 2020-11-20 2024-07-12 浙江大华技术股份有限公司 一种识别状态模型的训练方法、装置及计算机设备
CN112446860B (zh) * 2020-11-23 2024-04-16 中山大学中山眼科中心 一种基于迁移学习的糖尿病黄斑水肿自动筛查方法
CN113052229B (zh) * 2021-03-22 2023-08-29 武汉中旗生物医疗电子有限公司 一种基于心电数据的心脏病症分类方法及装置
CN113051368B (zh) * 2021-03-24 2023-09-22 北京百度网讯科技有限公司 双塔模型训练方法、检索方法、装置及电子设备
CN112967815A (zh) * 2021-04-08 2021-06-15 武汉爱尔眼科医院有限公司 一种基于干眼诊断综合***平台
CN113288091B (zh) * 2021-05-06 2023-10-03 广东工业大学 用于血压分类的模型训练方法、装置及可穿戴设备
CN113313110B (zh) * 2021-05-25 2024-02-13 北京易华录信息技术股份有限公司 一种车牌类型识别模型构建及车牌类型识别方法
WO2022252107A1 (zh) * 2021-06-01 2022-12-08 眼灵(上海)智能科技有限公司 一种基于眼部图像的疾病检测***及方法
CN113435310A (zh) * 2021-06-24 2021-09-24 中国科学技术大学 一种基于深度神经网络的视网膜数据采集处理方法及***
CN113470815A (zh) * 2021-06-30 2021-10-01 温州医科大学附属眼视光医院 基于ai技术的白内障患者视力损害程度评估***
CN113569655A (zh) * 2021-07-02 2021-10-29 广州大学 基于眼部色彩监控的红眼病患者识别***
CN113555077B (zh) * 2021-09-18 2022-01-11 北京大学第三医院（北京大学第三临床医学院） 疑似传染病预测方法及装置
CN114093473A (zh) * 2021-10-28 2022-02-25 广州爱尔眼科医院有限公司 一种眼科手术病人管理***
CN114451860A (zh) * 2022-01-27 2022-05-10 广东康软科技股份有限公司 一种基于深度学习的眼底病变诊断方法、***及设备
CN114512236A (zh) * 2022-04-18 2022-05-17 山东师范大学 一种阿尔兹海默症智能辅助诊断***
CN115147668B (zh) * 2022-09-06 2022-12-27 北京鹰瞳科技发展股份有限公司 疾病分类模型的训练方法、疾病分类的方法及相关产品
CN115514913B (zh) * 2022-09-16 2024-02-13 深圳市拓普智造科技有限公司 视频数据处理方法和装置、电子设备及存储介质
CN115910327B (zh) * 2022-12-22 2023-09-29 北京壹永科技有限公司 小样本癌症事件分析方法、装置、设备以及存储介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3065086A1 (en) * 2015-03-02 2016-09-07 Medizinische Universität Wien Computerized device and method for processing image data
US20160292856A1 (en) * 2015-04-06 2016-10-06 IDx, LLC Systems and methods for feature detection in retinal images
CN106682389A (zh) * 2016-11-18 2017-05-17 武汉大学 一种用于监控高血压引发的眼部病变健康管理***
CN106709254A (zh) * 2016-12-29 2017-05-24 天津中科智能识别产业技术研究院有限公司 一种医疗诊断机器人***
CN107358606A (zh) * 2017-05-04 2017-11-17 深圳硅基智能科技有限公司 用于识别糖尿病视网膜病变的人工神经网络及***
CN107437092A (zh) * 2017-06-28 2017-12-05 苏州比格威医疗科技有限公司 基于三维卷积神经网络的视网膜oct图像的分类算法
CN107680683A (zh) * 2017-10-09 2018-02-09 上海睦清视觉科技有限公司 一种ai眼部健康评估方法
CN107909621A (zh) * 2017-11-16 2018-04-13 深圳市唯特视科技有限公司 一种基于双生成对抗网络的医学图像合成方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6754380B1 (en) * 2003-02-14 2004-06-22 The University Of Chicago Method of training massive training artificial neural networks (MTANN) for the detection of abnormalities in medical images
US10803143B2 (en) * 2015-07-30 2020-10-13 Siemens Healthcare Gmbh Virtual biopsy techniques for analyzing diseases
GB201615051D0 (en) * 2016-09-05 2016-10-19 Kheiron Medical Tech Ltd Multi-modal medical image procesing
FR3055709B1 (fr) * 2016-09-05 2021-01-01 Essilor Int Methode de determination d'un equipement ophtalmique progressif de compensation visuelle personnalise pour un individu
CN106408562B (zh) * 2016-09-22 2019-04-09 华南理工大学 基于深度学习的眼底图像视网膜血管分割方法及***
KR101977645B1 (ko) * 2017-08-25 2019-06-12 주식회사 메디웨일 안구영상 분석방법
CN107644419A (zh) * 2017-09-30 2018-01-30 百度在线网络技术（北京）有限公司 用于分析医学影像的方法和装置
CN107680684B (zh) * 2017-10-12 2021-05-07 百度在线网络技术（北京）有限公司 用于获取信息的方法及装置

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3065086A1 (en) * 2015-03-02 2016-09-07 Medizinische Universität Wien Computerized device and method for processing image data
US20160292856A1 (en) * 2015-04-06 2016-10-06 IDx, LLC Systems and methods for feature detection in retinal images
CN106682389A (zh) * 2016-11-18 2017-05-17 武汉大学 一种用于监控高血压引发的眼部病变健康管理***
CN106709254A (zh) * 2016-12-29 2017-05-24 天津中科智能识别产业技术研究院有限公司 一种医疗诊断机器人***
CN107358606A (zh) * 2017-05-04 2017-11-17 深圳硅基智能科技有限公司 用于识别糖尿病视网膜病变的人工神经网络及***
CN107437092A (zh) * 2017-06-28 2017-12-05 苏州比格威医疗科技有限公司 基于三维卷积神经网络的视网膜oct图像的分类算法
CN107680683A (zh) * 2017-10-09 2018-02-09 上海睦清视觉科技有限公司 一种ai眼部健康评估方法
CN107909621A (zh) * 2017-11-16 2018-04-13 深圳市唯特视科技有限公司 一种基于双生成对抗网络的医学图像合成方法

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110840428A (zh) * 2019-11-29 2020-02-28 苏州大学 基于一维U-Net网络的无创血压估计方法
CN111048210A (zh) * 2019-12-31 2020-04-21 上海鹰瞳医疗科技有限公司 基于眼底图像评估疾病风险的方法及设备
CN111415741A (zh) * 2020-03-05 2020-07-14 北京深睿博联科技有限责任公司 基于隐式表观学习的乳腺x射线影像分类模型训练方法
CN111415741B (zh) * 2020-03-05 2023-09-26 北京深睿博联科技有限责任公司 基于隐式表观学习的乳腺x射线影像分类模型训练方法
CN111681756A (zh) * 2020-05-29 2020-09-18 吾征智能技术(北京)有限公司 基于痰液性状认知的疾病征兆预测***
CN112185564A (zh) * 2020-10-20 2021-01-05 福州数据技术研究院有限公司 一种基于结构化电子病历的眼科疾病预测方法和存储设备
CN112784801A (zh) * 2021-02-03 2021-05-11 紫东信息科技(苏州)有限公司 基于文本和图片的双模态胃部疾病分类方法及装置
CN112884729B (zh) * 2021-02-04 2023-08-01 北京邮电大学 基于双模态深度学习的眼底疾病辅助诊断方法和装置
CN112884729A (zh) * 2021-02-04 2021-06-01 北京邮电大学 基于双模态深度学习的眼底疾病辅助诊断方法和装置
CN113096804A (zh) * 2021-04-08 2021-07-09 武汉爱尔眼科医院有限公司 一种干眼症患者数据统计***
CN113435607A (zh) * 2021-06-09 2021-09-24 内蒙古大学 一种基于联邦学习的疾病筛查方法
CN113435607B (zh) * 2021-06-09 2023-08-29 内蒙古大学 一种基于联邦学习的疾病筛查方法
CN113421638A (zh) * 2021-06-22 2021-09-21 平安科技（深圳）有限公司 基于迁移学习的模型生成方法、装置和计算机设备
WO2023029510A1 (zh) * 2021-08-30 2023-03-09 康键信息技术（深圳）有限公司 基于人工智能的远程问诊方法、装置、设备及介质
CN113707301A (zh) * 2021-08-30 2021-11-26 康键信息技术（深圳）有限公司 基于人工智能的远程问诊方法、装置、设备及介质
CN115631367A (zh) * 2022-09-30 2023-01-20 中国医学科学院生物医学工程研究所 基于神经网络模型的眼科超声图像分类方法以及装置
CN115631367B (zh) * 2022-09-30 2024-03-05 中国医学科学院生物医学工程研究所 基于神经网络模型的眼科超声图像分类方法以及装置

Also Published As

Publication number Publication date
CN111656357A (zh) 2020-09-11
EP3783533A1 (en) 2021-02-24
US20210035689A1 (en) 2021-02-04
CN111656357B (zh) 2024-05-10
US11636340B2 (en) 2023-04-25

Similar Documents

Publication Publication Date Title
WO2019200535A1 (zh) 基于人工智能的眼科疾病诊断建模方法、装置及***
Rong et al. Surrogate-assisted retinal OCT image classification based on convolutional neural networks
Imran et al. Cataract detection and grading with retinal images using SOM-RBF neural network
KR102354980B1 (ko) 전안부 질환 진단 시스템 및 이를 이용한 진단 방법
Panda et al. Describing the structural phenotype of the glaucomatous optic nerve head using artificial intelligence
Uppamma et al. Deep learning and medical image processing techniques for diabetic retinopathy: a survey of applications, challenges, and future trends
Chakraborty et al. DCNN-based prediction model for detection of age-related macular degeneration from color fundus images
Bakır et al. Using transfer learning technique as a feature extraction phase for diagnosis of cataract disease in the eye
Rahhal et al. Detection and classification of diabetic retinopathy using artificial intelligence algorithms
Revathi et al. Particle Swarm Optimization based Detection of Diabetic Retinopathy using a Novel Deep CNN
Yang et al. Optimizing diabetic retinopathy detection with inception-V4 and dynamic version of snow leopard optimization algorithm
US20230157811A1 (en) Systems and methods for vitreous disease severity measurement
Pin et al. Retinal diseases classification based on hybrid ensemble deep learning and optical coherence tomography images
Vyas et al. A survey on automated eye disease detection using computer vision based techniques
Oh et al. Deep learning-based prediction of axial length using ultra-widefield fundus photography
Jolly et al. Eye Disease Detection using MobiNet
Zengin et al. Low-Resolution Retinal Image Vessel Segmentation
Selvathi Classification of ocular diseases using transfer learning approaches and glaucoma severity grading
Mani et al. An automated hybrid decoupled convolutional network for laceration segmentation and grading of retinal diseases using optical coherence tomography (OCT) images
US20240104731A1 (en) System for Integrated Analysis of Multi-Spectral Imaging and Optical Coherence Tomography Imaging
Ghosh et al. Finding correlation between diabetic retinopathy and diabetes during pregnancy based on computer-aided diagnosis: a review
Yadav et al. RD-Light-Net: Light Weight Network for Retinal Detachment Classification through Fundus Images
Ali et al. Classifying Three Stages of Cataract Disease using CNN
Nagaraj et al. Deep Learning Framework for Diabetic Retinopathy Diagnosis
US11355229B1 (en) System and method for generating an ocular dysfunction nourishment program

Legal Events

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

Ref document number: 18914959

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018914959

Country of ref document: EP

Effective date: 20201117