Image-based anti-VEGF indication identification method, device and electronic equipment

Through the data processing and model discrimination of optical coherent tomography images, the anti-VEGF indication was determined using the voting method, which solved the problem of inconsistent treatment caused by doctor interpretation differences, realized automated diagnosis in primary hospitals, improved treatment accuracy and efficiency.

CN112446876BActive Publication Date: 2024-11-15BEIJING DAHENG PUXIN MEDICAL TECH CO LTD
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202011464274.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-11
Publication Date
2024-11-15
Estimated Expiration
2040-12-11

AI Technical Summary

Technical Problem

The existing anti-VEGF treatment indications mainly rely on doctors' differences in the interpretation of optical coherent tomography images, resulting in inconsistent treatment plans, affecting the treatment effect, and requiring treatment in Grade A hospitals, increasing the patient's medical time and cost.

Method used

By obtaining the optical coherence tomography image to be detected, data processing and enhancement are performed, the trained discriminant model is input, and the anti-VEGF indication discrimination results are determined using the voting method, providing automated diagnostic suggestions and improving discriminant accuracy.

Benefits of technology

It realizes automated anti-VEGF indication judgment, improves diagnosis accuracy, and can be performed in grassroots hospitals, saves patients' medical time and costs, replaces doctors to perform image recognition, and improves medical experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112446876B_ABST
    Figure CN112446876B_ABST
Patent Text Reader

Abstract

The present application provides an anti-VEGF indication discrimination method, device and electronic device based on an image. The discrimination method comprises: obtaining an optical coherence tomography image to be detected; performing data processing on the optical coherence tomography image to obtain one or more test images; inputting the one or more test images into one or more trained discrimination models to obtain discrimination results of each test image by the one or more discrimination models; and determining the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method according to the discrimination results of each test image by the one or more discrimination models, so that a diagnostic suggestion can be automatically given to improve the discrimination accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an image-based anti-VEGF indication discrimination method, device, electronic equipment and computer storage medium. Background Art

[0002] Wet Age-related Macular Degeneration (wAMD) is an age-related degenerative disease that occurs in the macular area and causes fluid and lipid leakage and fibrous scarring due to the formation of choroidal neovascularization (CNV). It progresses rapidly and can cause severe visual impairment if not treated in time. In addition, the disease is complex, difficult to diagnose and treat, and requires a high level of professionalism from ophthalmologists.

[0003] Currently, intraocular injection of anti-vascular endothelial growth factor (VEGF) drugs is the first-line treatment for wAMD. Currently, the indication for anti-VEGF treatment is determined by doctors who analyze the optical coherence tomography (OCT) image set of the patient's macular area. If any characteristic lesions (representing choroidal neovascularization activity) exist in any OCT image in the image set, anti-VEGF treatment is considered necessary.

[0004] Different ophthalmologists have different interpretations of OCT images, resulting in inconsistent treatment plans, which in turn affects treatment outcomes. Conventional VEGF treatment requires going to a tertiary hospital for registration and treatment by a doctor. If a machine can replace a doctor to identify whether a patient needs anti-VEGF treatment, the process can be performed at a primary hospital, greatly saving the patient's time and cost of seeking medical treatment. Therefore, there is an urgent need for an automatic evaluation system for anti-VEGF indications to provide diagnostic recommendations and improve the accuracy of the judgment. Summary of the invention

[0005] The embodiment of the present application provides an image-based anti-VEGF indication discrimination method for automatically providing diagnostic suggestions and improving discrimination accuracy.

[0006] The present application embodiment provides an image-based anti-VEGF indication discrimination method, comprising:

[0007] acquiring an optical coherence tomography image to be detected;

[0008] Performing data processing on the optical coherence tomography image to obtain one or more test images;

[0009] Inputting the one or more test images into one or more trained discriminant models to obtain a discriminant result of the one or more discriminant models on each test image;

[0010] According to the discrimination results of each test image by the one or more discrimination models, the anti-VEGF indication discrimination result of the optical coherence tomography image is determined by a voting method.

[0011] In one embodiment, the performing data processing on the optical coherence tomography image to obtain one or more test images includes:

[0012] The optical coherence tomography image is flipped left and right, rotated, scaled, contrast changed, Gaussian noise changed, image cropped, or brightness changed.

[0013] In one embodiment, before inputting the one or more test images into the one or more trained discriminant models, the method further comprises:

[0014] Obtain sample images with marked characteristic lesions;

[0015] The sample image is used as the input of the target detection algorithm, and the parameters of the target detection algorithm are adjusted so that the lesion detection result output by the target detection algorithm is the same as the marked characteristic lesions, thereby obtaining the discrimination model.

[0016] In one embodiment, before inputting the one or more test images into the one or more trained discriminant models, the method further comprises:

[0017] Obtaining sample images labeled as wAMD positive and optical coherence tomography sample images labeled as having anti-VEGF indication;

[0018] The wAMD-positive sample images and the optical coherence tomography sample images with anti-VEGF indications are used as inputs of the classification algorithm, and the parameters of the classification algorithm are adjusted so that the classification results output by the classification algorithm are the same as the marked results, thereby obtaining the discrimination model.

[0019] In one embodiment, after obtaining the discriminant model, the method further includes:

[0020] Perform model distillation on the discriminant model to obtain a new discriminant model.

[0021] The present application also provides an image-based anti-VEGF indication discrimination device, comprising:

[0022] An image acquisition module, used for acquiring an optical coherence tomography image to be detected;

[0023] An image processing module, used for performing data processing on the optical coherence tomography image to obtain one or more test images;

[0024] An image discrimination module, used for inputting the one or more test images into one or more trained discrimination models to obtain discrimination results of the one or more discrimination models for each test image;

[0025] The result determination module is used to determine the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method according to the discrimination results of each test image by the one or more discrimination models.

[0026] In one embodiment, the image enhancement module is used to perform at least two changes of left-right flipping, rotation, image scaling, contrast change, Gaussian noise change, and brightness change on the optical coherence tomography image to obtain at least two test images.

[0027] In one embodiment, the device further comprises:

[0028] A sample acquisition module is used to acquire sample images with marked characteristic lesions;

[0029] The model training module is used to use the sample image as the input of the target detection algorithm, adjust the parameters of the target detection algorithm, make the lesion detection result output by the target detection algorithm the same as the marked characteristic lesion, and obtain the discrimination model

[0030] The present application also provides an electronic device, including:

[0031] processor;

[0032] a memory for storing processor-executable instructions;

[0033] Wherein, the processor is configured to execute the above-mentioned image-based anti-VEGF indication identification method.

[0034] An embodiment of the present application further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program can be executed by a processor to complete the above-mentioned method for distinguishing anti-VEGF indications of images.

[0035] The technical solution provided by the above-mentioned embodiment of the present application performs data processing on the optical coherence tomography image to obtain one or more test images, inputs the one or more test images into one or more trained discriminant models, and determines the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method based on the discrimination result of each test image by one or more discriminant models, thereby automatically giving diagnostic suggestions to improve the discrimination accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solution of the embodiments of the present application, the drawings required for use in the embodiments of the present application are briefly introduced below.

[0037] Figure 1 Schematic diagram of an application scenario of an anti-VEGF indication discrimination method based on an image provided in an embodiment of the present application;

[0038] Figure 2 is a schematic diagram of an electronic device provided in an embodiment of the present application;

[0039] Figure 3 It is a complete flow chart of the image-based anti-VEGF indication discrimination method provided in the embodiment of the present application;

[0040] Figure 4 It is a flow chart of a target detection algorithm model of an image-based anti-VEGF indication discrimination method provided in an embodiment of the present application;

[0041] Figure 5 It is a block diagram of a discrimination device of an anti-VEGF indication discrimination method based on an image provided in an embodiment of the present application;

[0042] Figure 6 yes Figure 5 Based on the corresponding embodiment, another embodiment of the present application shows a block diagram of a discrimination device for an anti-VEGF indication discrimination method based on an image;

[0043] Figure 7 yes Figure 6 Based on the corresponding embodiment, another embodiment of the present application shows a block diagram of a discrimination device for an image-based anti-VEGF indication discrimination method. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0045] In the description of the present application, use of suffixes such as 'module', 'part' or 'unit' used to indicate elements in the description is only for facilitating description of the present invention, and has no specific meaning by itself.

[0046] Figure 1 Schematic diagram of the application scenario of the image-based anti-VEGF indication discrimination method provided in the embodiment of the present application. Figure 1 As shown, the application scenario includes a client 110 and a server 120. The client 110 can send an optical coherence tomography image to be processed to the server 120, and the server 120 can perform data processing on the optical coherence tomography image to obtain one or more test images. After the image-based anti-VEGF indication discrimination model is trained, the server 120 can input one or more test images into the trained one or more discrimination models to obtain the discrimination results of the one or more discrimination models for each test image; according to the discrimination results of the one or more discrimination models for each test image, the anti-VEGF indication discrimination result of the optical coherence tomography image is determined by a voting method.

[0047] Figure 2 Schematic diagram of an electronic device provided in an embodiment of the present application. The electronic device 210 can be used as a server 120, and the electronic device 210 includes: a processor 230; a memory 220 for storing instructions executable by the processor 230; wherein the processor 230 is configured to execute the image-based anti-VEGF indication discrimination method provided in an embodiment of the present application. A communication interface 240 is used for communication between the electronic device 210 and an external device; and a communication bus 250 is used for communication between the memory 220, the processor 230 and the communication interface 240.

[0048] The memory 220 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The memory also stores multiple modules, which are respectively executed by the processor to complete the following steps of the image-based anti-VEGF indication discrimination method.

[0049] The embodiment of the present application further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program can be executed by the processor 230 to complete the following image-based anti-VEGF indication identification method.

[0050] Figure 3 Schematic diagram of the complete process of the image-based anti-VEGF indication identification method provided in the embodiment of the present application. Figure 3 As shown, the determination method may include the following steps S310 to S340.

[0051] Step S310: Acquire an optical coherence tomography image to be detected.

[0052] For the purpose of distinction, an optical coherence tomography image for which it is unknown whether anti-VEGF treatment is needed can be referred to as an optical coherence tomography image to be tested, or as an original optical coherence tomography image. The optical coherence tomography image to be tested, that is, the original optical coherence tomography image, is used in the prediction process; an optical coherence tomography image for which it is known whether anti-VEGF treatment is needed can be referred to as a sample image, and the sample image is used in the training process.

[0053] Step S320: performing data processing on the optical coherence tomography image to obtain one or more test images.

[0054] Data processing can be flipping the optical coherence tomography image left and right, rotating it (e.g., 15° left and right), scaling it, changing the contrast (e.g., [0.8, 1.2]), changing the Gaussian noise, cropping it, or changing the brightness. By transforming the optical coherence tomography image in a variety of enhancement methods, a variety of test images can be obtained. Without distinction, the image obtained after the optical coherence tomography image is processed is referred to as the test image here.

[0055] Step S330: input the one or more test images into one or more trained discriminant models to obtain discrimination results of the one or more discriminant models for each test image.

[0056] In one embodiment, one or more discriminant models can be trained in advance. The training process of the discriminant model can be shown in detail below. The same test image can be input into multiple discriminant models to obtain the discrimination results of multiple discriminant models for one test image. The discrimination results are used to indicate whether there is an anti-VEGF indication. Having an anti-VEGF indication can remind doctors to perform anti-VEGF treatment.

[0057] Step S340: Determine the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method according to the discrimination results of the one or more discrimination models for each test image.

[0058] For example, the voting method can be that when two or more models among multiple discriminant models judge that the same optical coherence tomography image needs anti-VEGF treatment, the image is finally judged to need anti-VEGF treatment. Among the discrimination results of multiple test images of the same optical coherence tomography image, at least two test images are considered to have anti-VEGF indications, then the optical coherence tomography image can be considered to have anti-VEGF indications. The result output by the discriminant model for the test image can be called the discrimination result. The judgment result of whether the optical coherence tomography image has anti-VEGF indications finally obtained by the voting method can be called the anti-VEGF indication discrimination result.

[0059] Figure 4 is a flow chart of the above-mentioned discriminant model provided in one embodiment of the present application. In one embodiment, the discriminant model can be obtained by training the target detection algorithm. Figure 4 As shown in Figure 2, the training process of the discriminant model includes:

[0060] Step S410: Acquire a sample image with marked characteristic lesions.

[0061] Since the criteria for judging the anti-VEGF indication are the presence of four characteristic lesions representing CNV activity, namely SRF (subretinal fluid), IRF (intraretinal fluid), PED (retinal pigment epithelial detachment) and SRHM (subretinal hyperreflective material), any OCT image showing any of the four characteristic lesions is considered to have an anti-VEGF indication. Therefore, the sample image here can be an optical coherence tomography image with the characteristic lesion location marked.

[0062] Step S420: using the sample image as input of a target detection algorithm, adjusting the parameters of the target detection algorithm so that the lesion detection result output by the target detection algorithm is the same as the marked characteristic lesions, and obtaining the discrimination model.

[0063] The target detection algorithm can output the lesion location representing CNV activity in the sample image. The training goal is to make the output lesion location as similar as possible to the annotated lesion location.

[0064] Figure 5 FIG. 1 is a flow chart of the above-mentioned discrimination model provided by another embodiment of the present application. In one embodiment, the discrimination model can be obtained by training a classification algorithm. Figure 5 As shown in Figure 2, the training process of the discriminant model includes:

[0065] Step S510: Acquire a sample image marked as wAMD positive and an optical coherence tomography sample image marked as having an anti-VEGF indication.

[0066] Sample images labeled wAMD positive are images in which wet age-related macular degeneration is not present.

[0067] Step S520: using the wAMD-positive sample images and the optical coherence tomography sample images with anti-VEGF indication as inputs to a classification algorithm, adjusting the parameters of the classification algorithm so that the classification results output by the classification algorithm are the same as the marked results, and obtaining the discrimination model.

[0068] The classification algorithm can classify the sample images into those with anti-VEGF indications and those without anti-VEGF indications. The training goal is to make the classification results of the sample images (eg, anti-VEGF indications or without anti-VEGF indications) as similar as possible to the actual annotation results.

[0069] In one embodiment, the discriminant model may also be implemented using an instance segmentation or semantic segmentation algorithm, and whether a specific anti-VEGF treatment indication is present is obtained based on the segmentation result.

[0070] In one embodiment, the sample image may be preprocessed first, and the data preprocessing includes two parts: data cleaning and data enhancement. The image obtained after the preprocessing is subjected to model training and testing.

[0071] Data cleaning refers to filtering out incorrectly labeled images. Data enhancement can flip, rotate, scale, change contrast, change Gaussian noise, crop, or change brightness of optical coherence tomography images. Data enhancement is used to improve the generalization ability of the model.

[0072] In one embodiment, a cross-validation method can be used to train multiple discriminant models. The backbone network adopts the Resnet series and the ResNeXt series, and the model parameters are initialized using the pre-trained model parameters of ImageNet. Since the semantic information of the images in ImageNet and the medical images is quite different, only the model parameters of the first stage are frozen during training, and the remaining parameters are tuned on the current data set. Finally, the top three discriminant models with the highest accuracy in the cross-validation process are selected for prediction. A predicted image is input into three discriminant models, and three results are output (whether anti-VEGF treatment is needed). Finally, the voting method (when two or more models believe that anti-VEGF treatment is needed, the final result is that anti-VEGF treatment is needed) is used to determine the final result. Among them, the backbone of the embodiment of the present application adopts Resnet101 with the highest accuracy.

[0073] After the discriminant model is trained, model distillation can be performed on the discriminant model to obtain a new discriminant model for deployment.

[0074] Model distillation refers to first training a complex model (teacher) with high accuracy, and then retraining a simple model. During the training process, the simple model (student) is used to fit the output of the complex model (teacher). In one embodiment, the output of the pre-trained discriminant model can be used as the known result, and the known result and the image to be detected can be merged into a new data set and put into another simple discriminant model for learning.

[0075] During the model deployment phase, the discriminant model needs to be deployed on the hardware device. At this time, factors such as computational complexity, model size, and speed need to be further considered. In order to obtain a better trade-off between model size and accuracy, the embodiment of the present application performs model distillation on the original integrated model (Resnet50 distilled into Resnet101), which reduces the hardware requirements for model deployment and improves the prediction speed at the cost of a slightly lower accuracy than the original.

[0076] Figure 6 is a block diagram of an anti-VEGF indication discrimination device based on an image provided in an embodiment of the present application. Figure 6 As shown, the device includes: an image acquisition module 610, an image processing module 620, an image identification module 630 and a result determination module 640.

[0077] An image acquisition module 610 is used to acquire an optical coherence tomography image to be detected;

[0078] An image processing module 620 is used to perform data processing on the optical coherence tomography image to obtain one or more test images;

[0079] An image discrimination module 630, configured to input the one or more test images into one or more trained discrimination models to obtain discrimination results of the one or more discrimination models for each test image;

[0080] The result determination module 640 is used to determine the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method according to the discrimination results of each test image by the one or more discrimination models.

[0081] In one embodiment, the data processing method includes: flipping the optical coherence tomography image left and right, rotating, scaling, changing the contrast, changing the Gaussian noise, cropping the image, or changing the brightness.

[0082] Figure 7 yes Figure 6 Based on the corresponding embodiment, another embodiment of the present application shows a block diagram of a discrimination device for an anti-VEGF indication discrimination method based on an image. Figure 7As shown, the device includes: a sample acquisition module 710 and a model training module 720.

[0083] The sample acquisition module 710 is used to acquire sample images with marked characteristic lesions;

[0084] The model training module 720 is used to use the sample image as the input of the target detection algorithm, adjust the parameters of the target detection algorithm, so that the lesion detection result output by the target detection algorithm is the same as the marked characteristic lesion, and obtain the discrimination model.

[0085] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned image-based anti-VEGF indication discrimination method, and will not be repeated here.

[0086] The technical solution provided in the present application can automatically give diagnostic suggestions, improve the accuracy of discrimination, and after practical application, can save patients' medical treatment time and improve their medical experience. This solution can use machines to replace doctors for image recognition, that is, to identify from the image whether the patient corresponding to the image needs anti-VEGF treatment, which greatly saves patients' medical treatment time and cost, and is a domestic first.

[0087] In several embodiments provided in the present application, the disclosed devices and methods may also be implemented in other ways. The device embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram may represent a module, a program segment or a part of a code, and a module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0088] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0089] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

Claims

1. An image-based anti-VEGF indication discrimination method, characterized in that: include: acquiring an optical coherence tomography image to be detected; Performing data processing on the optical coherence tomography image to obtain a plurality of test images; Inputting the plurality of test images into one or more trained discriminant models to obtain a discriminant result of the one or more discriminant models on each test image; Before inputting the plurality of test images into the one or more trained discriminant models, obtaining sample images with marked characteristic lesions; The sample image is used as the input of the target detection algorithm, and the parameters of the target detection algorithm are adjusted so that the lesion detection result output by the target detection algorithm is the same as the marked characteristic lesion, so as to obtain the discriminant model; a sample image marked as wAMD positive and an optical coherence tomography sample image marked as having anti-VEGF indication are obtained; the wAMD positive sample image and the optical coherence tomography sample image with anti-VEGF indication are used as the input of the classification algorithm, and the parameters of the classification algorithm are adjusted so that the classification result output by the classification algorithm is the same as the marked result, so as to obtain the discriminant model; wherein, the training process adopts the cross-validation method; the backbone network adopts the Resnet series and the ResNeXt series, and the model parameters are initialized with the pre-trained model parameters of ImageNet. During training, only the model parameters of the first stage are frozen, and the remaining parameters are tuned on the current data set; the top three discriminant models with the highest accuracy in the cross-validation process are selected as the discriminant models obtained by training; after the discriminant model is trained, the discriminant model is subjected to model distillation to obtain a new discriminant model; According to the discrimination results of each test image by the one or more discrimination models, the anti-VEGF indication discrimination result of the optical coherence tomography image is determined by a voting method; wherein the voting method is: among the discrimination results of multiple test images of the same optical coherence tomography image, if at least two test images are considered to have anti-VEFG indications, then the optical coherence tomography image has anti-VEFG indications.

2. The method according to claim 1, characterized in that The optical coherence tomography image is subjected to data processing to obtain a plurality of test images, including: The optical coherence tomography image is flipped left and right, rotated, scaled, contrast changed, Gaussian noise changed, image cropped, or brightness changed.

3. An image-based anti-VEGF indication discrimination device, characterized in that: include: An image acquisition module, used for acquiring an optical coherence tomography image to be detected; An image processing module, used for performing data processing on the optical coherence tomography image to obtain a plurality of test images; An image discrimination module, used for inputting the plurality of test images into one or more trained discrimination models to obtain discrimination results of the one or more discrimination models for each test image; A result determination module is used to determine the anti-VEGF indication discrimination result of the optical coherence tomography image by a voting method according to the discrimination results of each test image by the one or more discrimination models; wherein the voting method is: among the discrimination results of multiple test images of the same optical coherence tomography image, if at least two test images are considered to have anti-VEFG indication, then the optical coherence tomography image has anti-VEFG indication; A sample acquisition module, used for acquiring sample images with characteristic lesions marked before inputting the plurality of test images into the trained one or more discriminant models; acquiring sample images marked as wAMD positive and optical coherence tomography sample images marked as having anti-VEGF indications; A model training module is used to use the sample image as the input of the target detection algorithm, adjust the parameters of the target detection algorithm, so that the lesion detection result output by the target detection algorithm is the same as the marked characteristic lesion, and obtain the discriminant model; use the wAMD-positive sample image and the optical coherence tomography sample image with anti-VEGF indication as the input of the classification algorithm, adjust the parameters of the classification algorithm, so that the classification result output by the classification algorithm is the same as the marked result, and obtain the discriminant model; wherein, the training process adopts the cross-validation method; the backbone network adopts the Resnet series and the ResNeXt series, and the model parameters are initialized using the pre-trained model parameters of ImageNet. During training, only the model parameters of the first stage are frozen, and the remaining parameters are tuned on the current data set; the top three discriminant models with the highest accuracy in the cross-validation process are selected as the discriminant models obtained by training; after the discriminant model is trained, the discriminant model is subjected to model distillation to obtain a new discriminant model.

4. The device according to claim 3, characterized in that The data processing method includes: flipping the optical coherence tomography image left and right, rotating it, scaling it, changing its contrast, changing its Gaussian noise, cropping it, or changing its brightness.

5. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to execute the anti-VEGF indication identification method of an image as described in any one of claims 1-2.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program can be executed by a processor to complete the anti-VEGF indication identification method of an image according to any one of claims 1-2.

Citation Information

Patent Citations

  • Method for identifying unconventional cells in pathological section

    CN108346145A

  • Optical coherent tomographic image retinopathy intelligent testing system and testing method

    CN110010219A

  • Thyroid nuclear medicine image diagnosis method based on deep learning

    CN110021022A

  • Shareable semi-automatic diabetic fundus lesion labeling method

    CN110211087A

  • Ophthalmic disease classification and identification method and device based on random forest and medium

    CN110363226A