Eye fundus image focus segmentation and quantitative analysis method based on data enhancement

Abstract

The invention discloses a fundus image lesion segmentation and quantitative analysis method based on data enhancement, and belongs to the technical field of fundus, and the method comprises the following steps: S1, image preprocessing; s2, optic disc and macular positioning; s3, optic disc segmentation; s4, myopia bottom feature change recognition; and S5, leopard spot region feature quantification. The problems that accurate segmentation and quantitative analysis cannot be carried out on the focus of the fundus image, a high-quality diagnosis result and an accurate treatment scheme cannot be provided, the diagnosis accuracy is reduced, and the fundus treatment effect of a user is poor in the prior art are solved. The advanced artificial intelligence technology is utilized to deeply analyze and process the fundus image, myopia-related fundus changes such as leopard spot, diffuse chorioretinal atrophy and plaque-like chorioretinal atrophy can be accurately identified and quantitatively evaluated, and a visual heat map and a quantitative report are generated, so that myopia-related fundus changes can be accurately identified and quantitatively evaluated. A comprehensive and objective diagnosis basis is provided for doctors, and the doctors are assisted to formulate a treatment scheme more accurately.

Description

Technical Field

[0001] The present invention relates to the field of fundus technology, and in particular to a fundus image lesion segmentation and quantitative analysis method based on data enhancement. Background Art

[0002] Fundus images are increasingly used in ophthalmology clinics, mainly for diagnosing and monitoring the development of eye diseases. However, due to the complexity of eye diseases, lesions in fundus images are often unclear and changeable, making it difficult to accurately identify lesions in fundus images.

[0003] The Chinese patent with publication number CN103705207A discloses a fundus image system classified as a medical device for ophthalmic examination, which is composed of an electro-optical fundus speculum, a camera, and a computer. The camera is installed on the electro-optical fundus speculum to form an integrated whole. When in use, the transmission line of the camera is connected to the computer, so that the fundus image is transmitted to the computer through the camera and displayed on the screen of the display. It is suitable for electronic medical record systems and network diagnosis and treatment. However, the patent has the following defects:

[0004] Existing technologies cannot accurately segment and quantitatively analyze lesions in fundus images, cannot provide high-quality diagnostic results, reduce the accuracy of diagnosis, and cannot provide accurate treatment plans, resulting in poor fundus treatment effects for users. Summary of the invention

[0005] The purpose of the present invention is to provide a method for fundus image lesion segmentation and quantitative analysis based on data enhancement, which uses advanced artificial intelligence technology to perform in-depth analysis and processing of fundus images, and can accurately identify and quantitatively evaluate myopia-related fundus changes such as leopard spots, diffuse chorioretinal atrophy and patchy chorioretinal atrophy. By generating intuitive heat maps and quantitative reports, it provides doctors with a comprehensive and objective diagnostic basis, helps doctors formulate treatment plans more accurately, and solves the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for fundus image lesion segmentation and quantitative analysis based on data enhancement comprises the following steps:

[0008] S1. Image preprocessing:

[0009] Collect user's fundus photos based on retinal camera;

[0010] Perform ROI extraction, denoising, normalization and enhancement on the user's fundus photos;

[0011] Determine a preprocessed user fundus image;

[0012] S2, optic disc and macula positioning:

[0013] Localize the optic disc and macula based on deep learning object detection;

[0014] Train the optic disc and macula detection models;

[0015] Output the target detection frame based on the optic disc and macula detection model;

[0016] The center point of the detection frame is used as the center point of the optic disc and macula respectively;

[0017] S3, Video disc segmentation:

[0018] The center point of the optic disc obtained by optic disc positioning is taken as the origin;

[0019] Based on the computer vision attention mechanism, the user's fundus image is transformed into polar coordinates;

[0020] On the polar coordinate image, the edge detection operator is used to obtain the edge of the optic disc in polar coordinates;

[0021] Perform polar coordinate inverse transformation on the user's fundus image to obtain the optic disc edge in the rectangular coordinate system and perform fine segmentation on the optic disc;

[0022] S4. Identification of changes in myopic fundus features:

[0023] The deep learning semantic segmentation network TransUnet was used to establish the recognition models of leopard spot area, patchy atrophy area and diffuse atrophy area respectively;

[0024] The samples of the training set were input into the deep learning semantic segmentation network model to segment the leopard spot area, patchy atrophy area and diffuse atrophy area;

[0025] The test set is used to evaluate the segmentation accuracy of the deep learning semantic segmentation network model, and the accuracy, sensitivity, and specificity are calculated respectively;

[0026] S5. Quantification of leopard spot area features:

[0027] Based on the results of ROI and optic disc segmentation, the density of leopard spots within 4.5 mm with the optic disc center as the origin, the density of leopard spots within 1 mm, 3 mm, and 6 mm with the macula center as the origin, and the density of leopard spots in the upper and lower nasal and temporal areas and the upper and lower nasal, lower nasal, upper and lower temporal areas of the corresponding areas were calculated.

[0028] The calculation formula is as follows:

[0029]

[0030] Among them, S' is the area of ​​leopard spots in the region, S is the fundus area of ​​the corresponding region, and ρ is a dimensionless value, which indicates the proportion of leopard spots in the region.

[0031] Preferably, in S1, the user's fundus photo is subjected to region of interest ROI extraction, denoising, normalization and enhancement processing, and the following operations are performed:

[0032] Obtain user's fundus photo;

[0033] Based on image cropping and magnification technology, the region of interest (ROI) is extracted from the user's fundus photo;

[0034] Extract the area containing important information in the user's fundus photo, remove the irrelevant background in the user's fundus photo, and focus the analysis on the key parts of the user's fundus photo;

[0035] Based on advanced image denoising algorithms, denoise the user's fundus photos;

[0036] Replace the value of a point in the user's fundus photo with the median value of all points in a region of the point, so that pixels with relatively large grayscale differences from surrounding pixels are replaced with values ​​close to the surrounding pixel values, eliminating isolated noise points and removing noise from the user's fundus photo;

[0037] Based on histogram equalization and gamma correction technology, normalize the user's fundus photos;

[0038] Eliminate the brightness and contrast differences between the user's fundus photos to determine a consistent user's fundus image;

[0039] Based on image enhancement technology, enhance the user's fundus image;

[0040] The user's fundus image is contrast enhanced and sharpened to highlight the detailed features in the user's fundus image and make the lesion area more obvious.

[0041] Preferably, performing contrast enhancement processing on the user's fundus image includes:

[0042] Extracting the adjusted pixel grayscale value and the unadjusted pixel grayscale value corresponding to the noise point of the user's fundus image;

[0043] Extracting pixel grayscale values ​​corresponding to pixel points adjacent to noise points in the user's fundus image;

[0044] Obtaining a pixel grayscale adjustment degree coefficient corresponding to each noise point by using the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to each noise point in the user's fundus image and the pixel grayscale values ​​corresponding to the pixel points adjacent to the noise point in the user's fundus image;

[0045] The pixel grayscale adjustment coefficient corresponding to each noise point is obtained by the following formula:

[0046]

[0047] Wherein, L represents the pixel grayscale adjustment coefficient corresponding to each noise point; n represents the total number of pixel points adjacent to the noise point in the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the pixel grayscale value after adjustment and the pixel grayscale value before adjustment corresponding to the noise point; H z represents the pixel grayscale median value of n pixels adjacent to the noise point; g represents the adjustment factor, and the adjustment factor is obtained by the following formula:

[0048]

[0049] Wherein, g represents the adjustment factor; n represents the total number of pixel points adjacent to the noise point of the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to the noise point;

[0050] The contrast enhancement process is performed on the user fundus image by utilizing the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image.

[0051] Preferably, performing contrast enhancement processing on the user fundus image using pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image includes:

[0052] Extracting pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user's fundus image;

[0053] Extracting pixel points whose difference between the pixel grayscale value contained in the user fundus image and the average value of the adjusted pixel grayscale value corresponding to the noise point is lower than a preset grayscale difference threshold as target pixel points;

[0054] Obtaining a contrast adjustment coefficient using the pixel grayscale value corresponding to the target pixel point;

[0055] The contrast adjustment coefficient is obtained by the following formula:

[0056]

[0057] Where k represents the contrast adjustment coefficient; m represents the number of target pixels; H mi Represents the pixel gray value corresponding to the i-th target pixel; H sp Represents the average value of the adjusted pixel grayscale values ​​corresponding to all noise points; H sz Indicates the middle value of the adjusted pixel grayscale value corresponding to the noise point; H min Indicates the minimum grayscale value of the target pixel; H sm Indicates the maximum value of the adjusted pixel grayscale value corresponding to the noise point;

[0058] Obtaining a target contrast using the contrast adjustment coefficient and the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image;

[0059] The target contrast is obtained by the following formula:

[0060]

[0061] Among them, D m represents the target contrast; D represents the original grayscale value of the user's fundus image; a represents the total number of all noise points contained in the user's fundus image; L i represents the pixel grayscale adjustment coefficient corresponding to the i-th noise point; L z represents the middle value of the pixel grayscale adjustment coefficient corresponding to n noise points; k represents the contrast adjustment coefficient;

[0062] The contrast enhancement adjustment is performed on the user fundus image according to the target contrast.

[0063] Preferably, in S2, the optic disc and macula detection model is trained, and the following operations are performed:

[0064] According to the user's requirements for locating the optic disc and macula in the fundus image, annotated samples of the optic disc and macula are collected, and annotated optic disc samples and macula samples are determined;

[0065] The network model of single-step detection SSD of target detection is selected, in which the backbone network in the network structure uses ResNet-50 network;

[0066] The network model of single-step detection SSD of selected target detection is trained based on the labeled optic disc samples and macula samples, and the network model parameters are continuously optimized;

[0067] The optic disc and macula detection models were determined.

[0068] Preferably, in S2, the target detection frame is output based on the optic disc and macula detection model, and the following operations are performed:

[0069] Obtain the optic disc and macula detection model;

[0070] Input the preprocessed user fundus image into the optic disc and macula detection model;

[0071] Perform target detection on the pre-processed user fundus image based on the optic disc and macula detection model, so that the target area can be accurately detected at different scales;

[0072] Based on the optic disc and macula detection model, the target detection frame is output, and the center point of each area is extracted as the center point of the optic disc and macula.

[0073] Preferably, in S3, the video disc is finely segmented by performing the following operations:

[0074] Polar coordinate conversion: Taking the center of the optic disc as the origin, the user's fundus image is converted into polar coordinates;

[0075] Edge detection: Use edge detection operators to identify the edges of the optic disc;

[0076] Edge delineation: The identified edges are converted back into Cartesian coordinates for delineation, and the optic disc area is accurately segmented by delineating the edges.

[0077] Preferably, in S4, the leopard spot area, patchy atrophy area and diffuse atrophy area are segmented by performing the following operations:

[0078] According to the recognition requirements of myopic fundus features, the samples were labeled by two doctors;

[0079] One of the physicians provided the initial annotations, and the other physician reviewed and revised them;

[0080] Among them, the labeled samples include leopard spot areas, patchy atrophy areas, and diffuse atrophy areas;

[0081] The labeled samples are divided into training set and test set in a ratio of 7:3;

[0082] During training, the samples of the training set are input into the deep learning semantic segmentation network model;

[0083] Among them, the deep learning semantic segmentation network model uses the Transformer structure in the encoder to extract high-level features from the input user's fundus image;

[0084] In the decoding stage, the extracted high-level features are restored to the resolution of the input image by upsampling;

[0085] The deep learning semantic segmentation network model outputs a confidence probability map corresponding to the extracted high-level features;

[0086] Threshold segmentation is performed based on the confidence probability map to segment the leopard spot area, patchy atrophy area and diffuse atrophy area.

[0087] Preferably, in S4, the accuracy, sensitivity and specificity are calculated by performing the following operations:

[0088] Taking image pixels as the smallest analysis unit, the accuracy, sensitivity and specificity of leopard spot area, patchy atrophy area and diffuse atrophy area were calculated respectively;

[0089] Among them, the accuracy, sensitivity, and specificity of the leopard spot region were 0.9766, 0.9551, and 0.9801, respectively;

[0090] Among them, the accuracy, sensitivity, and specificity of patchy atrophy area were 0.9993, 0.9795, and 0.9995, respectively;

[0091] Among them, the accuracy, sensitivity and specificity of diffuse atrophy area were 0.997, 0.978 and 0.998 respectively.

[0092] Preferably, in S5, after the leopard spot density is calculated and quantified, a user fundus quantitative report is generated, wherein the user fundus quantitative report includes detailed information on the area, shape, and severity of the lesion area. Based on the user fundus quantitative report, the doctor is assisted in formulating a personalized treatment plan for the user and performing personalized treatment on the user.

[0093] Compared with the prior art, the present invention has the following beneficial effects:

[0094] The present invention uses advanced artificial intelligence technology to perform in-depth analysis and processing of fundus images. It can accurately identify and quantitatively evaluate myopia-related fundus changes such as leopard spots, diffuse chorioretinal atrophy, and patchy chorioretinal atrophy. By generating intuitive heat maps and quantitative reports, it provides doctors with a comprehensive and objective diagnostic basis, helping doctors to formulate treatment plans more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 It is a flow chart of the method for segmenting and quantitatively analyzing fundus image lesions of the present invention;

[0096] Figure 2 A lesion extraction flow chart of the method for segmenting and quantitatively analyzing lesions in fundus images of the present invention;

[0097] Figure 3 This is a schematic diagram of selected characteristic indicators of pathological myopia in the method for segmenting and quantitatively analyzing fundus image lesions of the present invention. DETAILED DESCRIPTION

[0098] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0099] In order to solve the existing problem that the fundus image lesions cannot be accurately segmented and quantitatively analyzed, high-quality diagnostic results cannot be provided, the accuracy of diagnosis is reduced, accurate treatment plans cannot be provided, and the treatment effect of the fundus is poor, please refer to Figure 1-Figure 3 , this embodiment provides the following technical solutions:

[0100] A method for fundus image lesion segmentation and quantitative analysis based on data enhancement comprises the following steps:

[0101] S1. Image preprocessing:

[0102] Collect user's fundus photos based on retinal camera;

[0103] Perform region of interest (ROI) extraction, denoising, normalization and enhancement on user’s fundus photos;

[0104] It can eliminate non-fundus structures in the user's fundus photos, improve the clarity of the edges of internal features, and reduce the differences between images. Such preprocessing lays a solid foundation for subsequent advanced image analysis;

[0105] Determine a preprocessed user fundus image;

[0106] It should be noted that, with the connection between the retinal camera and the cloud server, fundus images can be obtained instantly and online diagnosis can be performed, providing convenient and efficient telemedicine services for patients. No matter where the patient is, as long as there is network conditions, they can enjoy professional fundus examination services, which greatly alleviates the problem of tight medical resources and improves the accessibility and convenience of medical services, including:

[0107] Real-time image transmission: Through real-time connection with the retinal camera and cloud server, the patient's fundus image can be instantly obtained and uploaded to the cloud server for analysis and processing. In this way, no matter where the patient is, as long as there is network conditions, the fundus examination can be carried out in real time to obtain professional diagnosis results;

[0108] Online diagnosis: It can analyze and diagnose uploaded fundus images in real time and return the diagnosis results to doctors and patients. Doctors can view the diagnosis results through the online platform and make further diagnosis and treatment suggestions. Patients can also obtain diagnosis results through the online platform and understand their eye health status in a timely manner.

[0109] Telemedicine service: Through telemedicine service, patients can get professional fundus examination service without going to the hospital. It is especially suitable for remote areas with scarce medical resources. Telemedicine service not only improves the accessibility and convenience of medical services, but also reduces the travel time and cost of patients, thus improving the medical experience of patients.

[0110] Electronic medical record management: Supports the storage of diagnostic results in the electronic medical record system, making it convenient for doctors to check and compare the changes in the patient's fundus at any time. The application of electronic medical records improves the continuity and accuracy of medical services and provides strong support for long-term eye disease management and follow-up.

[0111] In this embodiment, the user's fundus photo is subjected to ROI extraction, denoising, normalization and enhancement processing, and the following operations are performed:

[0112] Obtain user's fundus photo;

[0113] Based on image cropping and magnification technology, the region of interest (ROI) is extracted from the user's fundus photo;

[0114] Extract the area containing important information in the user's fundus photo, remove the irrelevant background in the user's fundus photo, and focus the analysis on the key parts of the user's fundus photo;

[0115] Based on advanced image denoising algorithms, denoise the user's fundus photos;

[0116] Replace the value of a point in the user's fundus photo with the median value of all points in a range of the point, so that pixels with relatively large grayscale values ​​of surrounding pixels are replaced with values ​​close to the surrounding pixel values, eliminate isolated noise points, remove noise in the user's fundus photo, improve the clarity and contrast of the user's fundus photo, and ensure that the details of the lesion area can be accurately captured;

[0117] Based on histogram equalization and gamma correction technology, normalize the user's fundus photos;

[0118] Eliminate the brightness and contrast differences between the user's fundus photos to determine a consistent user's fundus image;

[0119] Based on image enhancement technology, enhance the user's fundus image;

[0120] The user's fundus image is contrast enhanced and sharpened to highlight the detailed features in the user's fundus image, making the lesion area more obvious, providing a reliable basis for subsequent automated recognition and diagnosis.

[0121] Specifically, contrast enhancement processing is performed on the user's fundus image, including:

[0122] Extracting the adjusted pixel grayscale value and the unadjusted pixel grayscale value corresponding to the noise point of the user's fundus image;

[0123] Extracting pixel grayscale values ​​corresponding to pixel points adjacent to noise points in the user's fundus image;

[0124] Obtaining a pixel grayscale adjustment degree coefficient corresponding to each noise point by using the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to each noise point in the user's fundus image and the pixel grayscale values ​​corresponding to the pixel points adjacent to the noise point in the user's fundus image;

[0125] The pixel grayscale adjustment coefficient corresponding to each noise point is obtained by the following formula:

[0126]

[0127] Wherein, L represents the pixel grayscale adjustment coefficient corresponding to each noise point; n represents the total number of pixel points adjacent to the noise point in the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the pixel grayscale value after adjustment and the pixel grayscale value before adjustment corresponding to the noise point; H z represents the pixel grayscale median value of n pixels adjacent to the noise point; g represents the adjustment factor, and the adjustment factor is obtained by the following formula:

[0128]

[0129] Wherein, g represents the adjustment factor; n represents the total number of pixel points adjacent to the noise point of the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to the noise point;

[0130] The contrast enhancement process is performed on the user fundus image by utilizing the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image.

[0131] The technical effect of the above technical solution is as follows: the solution first focuses on the noise points in the user's fundus image, and provides basic data for subsequent processing by extracting the pixel grayscale values ​​of these noise points before and after adjustment, as well as the pixel grayscale values ​​of the pixels adjacent to them. This targeted processing method helps to more accurately identify and process the noise in the image, avoiding unnecessary interference to the image as a whole. By calculating the pixel grayscale adjustment degree coefficient L corresponding to each noise point, the solution can dynamically determine the grayscale adjustment degree of each noise point based on the grayscale value difference between the noise point and its adjacent pixels. This dynamic adjustment mechanism makes the contrast enhancement processing more flexible and accurate, and can better adapt to the characteristics of different fundus images.

[0132] The introduction of the adjustment factor g further enhances the flexibility and adaptability of the scheme. By comprehensively considering the grayscale value difference between the noise point and its adjacent pixels, as well as the total number of adjacent pixels n, the adjustment factor g can more comprehensively reflect the local characteristics of the fundus image, thereby guiding the calculation of the pixel grayscale adjustment coefficient, making the contrast enhancement processing more reasonable and effective. Through contrast enhancement processing, the scheme can significantly improve the quality of the user's fundus image. On the one hand, by reducing the interference of noise points, the image is clearer and more accurate; on the other hand, by enhancing the contrast of the image, the fundus structure is more prominent and easier to identify, which is helpful for subsequent medical diagnosis and analysis.

[0133] In summary, this technical solution achieves more precise and effective contrast enhancement processing of user fundus images through targeted noise point processing, dynamic adjustment of pixel grayscale, introduction of adjustment factor g, and improvement of image quality. This is of great significance for improving the recognition accuracy and diagnosis efficiency of fundus images.

[0134] Specifically, performing contrast enhancement processing on the user fundus image by using the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image includes:

[0135] Extracting pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user's fundus image;

[0136] Extracting pixel points whose difference between the pixel grayscale value contained in the user fundus image and the average value of the adjusted pixel grayscale value corresponding to the noise point is lower than a preset grayscale difference threshold as target pixel points;

[0137] Obtaining a contrast adjustment coefficient using the pixel grayscale value corresponding to the target pixel point;

[0138] The contrast adjustment coefficient is obtained by the following formula:

[0139]

[0140] Where k represents the contrast adjustment coefficient; m represents the number of target pixels; H mi Represents the pixel gray value corresponding to the i-th target pixel; H sp Represents the average value of the adjusted pixel grayscale values ​​corresponding to all noise points; H sz Indicates the middle value of the adjusted pixel grayscale value corresponding to the noise point; H min Indicates the minimum grayscale value of the target pixel; H sm Indicates the maximum value of the adjusted pixel grayscale value corresponding to the noise point;

[0141] Obtaining a target contrast using the contrast adjustment coefficient and the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image;

[0142] The target contrast is obtained by the following formula:

[0143]

[0144] Among them, D m represents the target contrast; D represents the original grayscale value of the user's fundus image; a represents the total number of all noise points contained in the user's fundus image; L i represents the pixel grayscale adjustment coefficient corresponding to the i-th noise point; L z represents the middle value of the pixel grayscale adjustment coefficient corresponding to n noise points; k represents the contrast adjustment coefficient;

[0145] The contrast enhancement adjustment is performed on the user fundus image according to the target contrast.

[0146] The technical effect of the above technical solution is: by extracting the pixel grayscale adjustment degree coefficient corresponding to all noise points in the user's fundus image, the solution can perform refined contrast adjustment for each noise point. This refined processing helps to reduce the impact of noise on image quality while maintaining important details and features in the fundus image. By setting the grayscale difference threshold, the pixel points with a grayscale value close to the adjusted target noise point are screened out as the target pixel points. This can more accurately reflect the grayscale characteristics of the local area in the fundus image and provide a reliable basis for the subsequent calculation of the contrast adjustment coefficient.

[0147] The calculation of the contrast adjustment coefficient k comprehensively considers multiple factors such as the grayscale value of the target pixel, the average grayscale value of the noise point after adjustment, the median value, the maximum value and the minimum value. This dynamic adjustment mechanism makes the contrast enhancement processing more flexible and adaptable, and can automatically adjust the contrast according to the specific situation of the fundus image to achieve the best visual effect. The target contrast is calculated by combining the original grayscale value of the user's fundus image, the pixel grayscale adjustment degree coefficient of the noise point and the contrast adjustment coefficient. This calculation method not only considers the overall grayscale characteristics of the fundus image, but also takes into account the contrast requirements of the local area, making the adjusted image more natural and realistic. After the above contrast enhancement processing, the quality of the user's fundus image is significantly improved. The details in the image are clearer and the contrast is more vivid, which helps doctors better observe and diagnose fundus lesions. At the same time, the scheme also has certain versatility and scalability, and can be applied to different types of fundus image processing and diagnosis tasks.

[0148] In summary, this technical solution achieves effective contrast enhancement processing of user fundus images through steps such as refined contrast adjustment, target pixel point screening, dynamic contrast adjustment coefficient calculation, and target contrast calculation. This processing method not only improves image quality, but also provides strong support for accurate diagnosis of fundus lesions.

[0149] S2, optic disc and macula positioning:

[0150] Localize the optic disc and macula based on deep learning object detection;

[0151] According to the user's requirements for locating the optic disc and macula in the fundus image, annotated samples of the optic disc and macula are collected, and annotated optic disc samples and macula samples are determined;

[0152] The network model of single-step detection SSD of target detection is selected. The backbone network in the network structure uses the ResNet-50 network, which can quickly and efficiently identify and locate the optic disc and macula in the user's fundus image;

[0153] The network model of single-step detection SSD of selected target detection is trained based on the labeled optic disc samples and macula samples, and the network model parameters are continuously optimized. The model parameters are continuously optimized through training with large-scale fundus image data sets to improve the accuracy and robustness of detection.

[0154] Determine the optic disc and macula detection model;

[0155] Obtain the optic disc and macula detection model;

[0156] Input the preprocessed user fundus image into the optic disc and macula detection model;

[0157] Based on the optic disc and macula detection model, the pre-processed user fundus image is used for target detection, so that the target area can be accurately detected at different scales. The multi-scale feature map is used for target detection to ensure that the target area can be accurately detected at different scales, thereby improving the accuracy and stability of detection.

[0158] Output the target detection frame based on the optic disc and macula detection model, and extract the center point of each area as the center point of the optic disc and macula;

[0159] It should be noted that in fundus photographs, the optic disc and macula are important reference points. In order to accurately identify these areas, deep learning target detection technology is used, especially the SSD (Single Shot Detection) network model. Through training, the model can efficiently locate the position of the optic disc and macula in fundus photographs, providing key reference information for subsequent fundus analysis.

[0160] S3, Video disc segmentation:

[0161] The center point of the optic disc obtained by optic disc positioning is taken as the origin;

[0162] Based on the computer vision attention mechanism, the user's fundus image is transformed into polar coordinates;

[0163] On the polar coordinate image, the edge detection operator is used to obtain the edge of the optic disc in polar coordinates;

[0164] Perform polar coordinate inverse transformation on the user's fundus image to obtain the optic disc edge in the rectangular coordinate system and perform fine segmentation on the optic disc;

[0165] In this embodiment, the video disc is finely segmented by performing the following operations:

[0166] Polar coordinate conversion: The user's fundus image is converted into polar coordinates with the center of the optic disc as the origin, which simplifies the identification and segmentation of the optic disc edge;

[0167] Edge detection: Use edge detection operators to identify the edge of the optic disc, such as Canny edge detection, to identify the edge of the optic disc and improve the accuracy of edge extraction;

[0168] Edge delineation: The identified edges are converted back into Cartesian coordinates for delineation, and the optic disc area is accurately segmented by delineating the edges.

[0169] It should be noted that after determining the position of the optic disc, the attention mechanism in computer vision is further used to achieve accurate segmentation of the optic disc. The fundus image is converted into polar coordinates, and the edge of the optic disc is identified using an edge detection operator, and then converted back to Cartesian coordinates for outlining, thereby achieving accurate extraction of the optic disc area. This plays an important auxiliary role in the subsequent analysis of myopic fundus changes.

[0170] S4. Identification of changes in myopic fundus features:

[0171] The deep learning semantic segmentation network TransUnet was used to establish the recognition models of leopard spot area, patchy atrophy area and diffuse atrophy area respectively. The deep learning semantic segmentation network TransUnet structure was used to enhance the global feature expression ability and ensure the effective extraction of large-area lesions.

[0172] According to the recognition requirements of myopic fundus features, the samples were labeled by two doctors;

[0173] One of the physicians provided the initial annotations, and the other physician reviewed and revised them;

[0174] Among them, the labeled samples include leopard spot areas, patchy atrophy areas, and diffuse atrophy areas;

[0175] The number of labeled samples is 1160, 2700, and 2780 respectively, and they are divided into training set and test set in a ratio of 7:3;

[0176] During training, the samples of the training set are input into the deep learning semantic segmentation network model;

[0177] Among them, the deep learning semantic segmentation network model uses the Transformer structure in the encoder to extract high-level features from the input user's fundus image;

[0178] In the decoding stage, the extracted high-level features are restored to the resolution of the input image by upsampling;

[0179] The deep learning semantic segmentation network model outputs a confidence probability map corresponding to the extracted high-level features;

[0180] Threshold segmentation is performed based on the confidence probability map to segment the leopard spot area, patchy atrophy area, and diffuse atrophy area.

[0181] The test set was used to evaluate the segmentation accuracy of the deep learning semantic segmentation network model. The image pixel was used as the smallest analysis unit to calculate the accuracy, sensitivity, and specificity of the leopard spot area, patchy atrophy area, and diffuse atrophy area.

[0182] Specifically, pathological myopia is an eye disease characterized by progressive deepening of refraction, continuous growth of eye axis, and progressive damage to ocular contents and retinal choroidal tissue, causing visual dysfunction. It occurs at an early age and progresses rapidly. It may cause serious complications such as macular atrophy, retinal pigment epithelial atrophy, choroidal neovascularization and subretinal hemorrhage. It is one of the main causes of blindness in my country.

[0183] Among them, leopard-like fundus is a fundus lesion characterized by increased visibility of large choroidal vessels around the fovea and arch vessels, accompanied by thinning of the retinal epithelium and choroidal capillaries. It is characterized by thinning of the retinal pigment epithelium and choroidal capillaries. Multiple factors are associated with the occurrence of leopard-like fundus lesions, including age, myopia, early age-related macular degeneration, glaucoma, and body mass index. The characteristic description includes increased visibility of large choroidal vessels around the fovea and arch vessels, thinning of the retinal epithelium and choroidal capillaries. Factors leading to leopard-like fundus lesions include aging, myopia development, early age-related macular degeneration, glaucoma, and body mass index. Through fundus image analysis, the leopard-like lesion area is identified and quantitatively evaluated to provide detailed diagnostic information.

[0184] Among them, diffuse chorioretinal atrophy is a relatively early fundus lesion in myopic atrophic maculopathy, which manifests as yellow-white changes with unclear boundaries in the posterior pole of the eyeball, and blurred choroidal blood vessels in the atrophic part. As the lesion progresses, the atrophic lesions can be widely distributed throughout the posterior pole of the eyeball, which is one of the fundus lesions that cause myopic visual function damage; diffuse atrophy is an important critical feature for diagnosing pathological myopia. Early intervention and treatment of this disease are of great significance for delaying the progression of pathological myopia and reducing visual function damage. It is characterized by yellow-white changes with unclear boundaries in the posterior pole of the eyeball, blurred choroidal blood vessels in the atrophic part, and lesions can be widely distributed throughout the posterior pole of the eyeball. Diffuse atrophy is a relatively early lesion in myopic atrophic maculopathy. As the disease progresses, the atrophic lesions gradually expand, leading to visual function damage. Through fundus image analysis, diffuse atrophy areas can be identified and quantitatively evaluated to provide detailed diagnostic information;

[0185] Among them, patchy chorioretinal atrophy often manifests as gray-white lesions with clear boundaries, which are between one and several choroidal lobules in size and often appear in the macular area or around the optic disc; the anatomical structure of patchy atrophy is essentially the loss of retinal pigment epithelium and most of the choroid, and is characterized by gray-white lesions with clear boundaries, which usually appear in the macular area or around the optic disc, and are between one and several choroidal lobules in size. Patchy atrophy is caused by the loss of retinal pigment epithelium and most of the choroid, and is common in patients with high myopia. Through fundus image analysis, the patchy atrophy area can be identified and quantitative evaluation can be performed to provide detailed diagnostic information.

[0186] It should be noted that the calculation results of the leopard spot area, patchy atrophy area and diffuse atrophy area are shown in Table 1:

[0187] Table 1: Calculation results of leopard spot area, patchy atrophy area and diffuse atrophy area

[0188]

[0189] Among them, the accuracy, sensitivity, and specificity of the leopard spot region were 0.97678, 0.95506, and 0.98006, respectively;

[0190] Among them, the accuracy, sensitivity, and specificity of patchy atrophy area were 0.99929, 0.97951, and 0.99951, respectively;

[0191] Among them, the accuracy, sensitivity and specificity of diffuse atrophy area were 0.99627, 0.97734 and 0.99764 respectively.

[0192] It should be noted that after the leopard spot density is calculated and quantified, a user fundus quantitative report is generated, where the user fundus quantitative report includes detailed information on the area, shape, and severity of the lesion area. Based on the user fundus quantitative report, the doctor can develop a personalized treatment plan for the user and provide personalized treatment to the user.

[0193] It should be noted that the most advanced deep learning algorithm is used to perform detailed analysis and processing of fundus images, which can automatically identify the diseased areas in the fundus images and perform quantitative assessment of these areas, including measuring indicators such as the area, shape and severity of the lesions. Through these analyses, detailed lesion information can be provided to help doctors better understand the patient's eye disease. An intuitive heat map can be generated to show the distribution and severity of the lesion area. The heat map can clearly show the specific location and range of the lesion, helping doctors to quickly locate the lesion area and make more accurate diagnosis and treatment, thereby improving the accuracy and effectiveness of treatment. It can significantly improve the accuracy and efficiency of fundus image analysis, process a large number of fundus images in a short time, provide high-quality diagnostic results, reduce the workload of doctors, and improve the accuracy of diagnosis.

[0194] Specifically, the application of artificial intelligence fundus quantitative technology can help primary ophthalmologists to more accurately detect and diagnose fundus lesions and improve the ability to prevent and control eye diseases. At the same time, the construction of an eye health big data center with the help of the AI ​​platform can realize the integration, analysis and sharing of eye disease information, provide strong data support for primary medical institutions, and promote the intelligent and precise prevention and control of eye diseases. Specifically, it includes:

[0195] Primary screening technology: It can provide primary medical institutions with advanced eye disease screening technology. Through regular fundus examinations, it can help primary doctors to timely detect and diagnose fundus lesions and improve the quality and level of primary medical services.

[0196] Improved prevention and control capabilities: Through the application of artificial intelligence technology, it can automatically identify and quantify fundus lesions, provide detailed lesion information, help grassroots doctors improve their eye disease prevention and control capabilities, and formulate more scientific prevention and control measures;

[0197] Big data support: With the help of AI platform, an eye health big data center is built to integrate and analyze a large amount of eye disease information and provide comprehensive data support for primary medical institutions. The results of big data analysis can not only help doctors better understand the patient's condition, but also provide a scientific basis for the formulation of public health policies;

[0198] Intelligent prevention and treatment: Through intelligent eye disease screening and prevention and control systems, primary medical institutions can achieve early detection, early diagnosis and early treatment of eye diseases, improve the efficiency and quality of prevention and control work, and reduce the incidence of eye diseases and blindness rates.

[0199] At the same time, in response to the needs of chronic disease screening systems for children, adolescents and remote areas, regular screening and monitoring can timely detect and intervene in the development of eye diseases such as myopia, reduce the risk of vision loss, and at the same time, by building a complete screening system, long-term tracking and management of chronic disease patients can be achieved, providing strong support for the formulation of personalized prevention and treatment plans; specifically including:

[0200] Regular screening and monitoring: It can conduct regular fundus examination and monitoring on children, adolescents and patients in remote areas, timely detect and diagnose fundus lesions, especially myopia-related eye diseases, and provide timely intervention measures.

[0201] Personalized prevention and treatment plan: Through detailed lesion analysis and diagnostic reports, it can provide doctors with comprehensive diagnostic basis, assist doctors in formulating personalized prevention and treatment plans, improve prevention and treatment effects, and reduce the incidence and blindness rate of eye diseases.

[0202] Chronic disease management: Support long-term tracking and management of patients with chronic diseases. Through regular examinations and follow-ups, timely understand changes in patients' conditions, adjust treatment plans, and provide continuous medical services.

[0203] Improve screening coverage: Through promotion and application in remote areas and primary medical institutions, the coverage of eye disease screening can be significantly improved, ensuring that more patients can detect and treat eye diseases in a timely manner and improving their eye health.

[0204] At the same time, with the help of AI platform, the integration and sharing of eye disease information can be realized, and comprehensive eye health management services can be provided for medical institutions at all levels. By building a hierarchical diagnosis and treatment system, the reasonable diversion and effective treatment of eye disease patients can be realized, the utilization efficiency of medical resources and service quality can be improved, which will help to improve the overall level of eye disease prevention and treatment in my country and bring good news to the majority of eye disease patients. Specifically, it includes:

[0205] Information integration and sharing: Through the AI ​​platform, comprehensive integration and sharing of eye disease information can be achieved, providing detailed eye disease data support for medical institutions at all levels and promoting the optimal allocation of medical resources;

[0206] Hierarchical diagnosis and treatment system: It can reasonably divert patients to different levels of medical institutions for treatment according to the severity of their condition, ensuring that patients can obtain the most appropriate medical services and improve treatment effects;

[0207] Resource utilization efficiency: Through the hierarchical diagnosis and treatment system, the utilization efficiency of medical resources can be optimized, the waste of medical resources can be avoided, and the overall quality and efficiency of medical services can be improved;

[0208] Comprehensive management services: Provide comprehensive eye health management services, including prevention, screening, diagnosis, treatment and follow-up, to ensure that patients can receive continuous and comprehensive medical services and improve patient satisfaction and treatment effects.

[0209] It should be noted that in order to support model training and verification, a high-quality fundus image database was built based on the China Consortium for Research on High Myopia (CHARM). The database covers fundus images of different ages, genders, demographic characteristics, and various fundus camera types, and the data are labeled by multiple fundus experts. This rich data set provides a solid foundation for subsequent research and applications; specifically, it includes:

[0210] Data collection: Collect fundus images of different ages, genders, demographic characteristics, and fundus camera types to ensure data diversity and coverage;

[0211] Data labeling: Data labeling is performed by multiple ophthalmologists to ensure the accuracy and reliability of labeling and provide high-quality data support for model training;

[0212] Database management: Build a fundus image database, manage and maintain the data, ensure the integrity and availability of the data, and provide a solid foundation for subsequent research and application.

[0213] S5. Quantification of leopard spot area features:

[0214] Based on the results of ROI and optic disc segmentation, the density of leopard spots within 4.5 mm with the optic disc center as the origin, the density of leopard spots within 1 mm, 3 mm, and 6 mm with the macula center as the origin, and the density of leopard spots in the upper and lower nasal and temporal areas and the upper and lower nasal, lower nasal, upper and lower temporal areas of the corresponding areas were calculated.

[0215] The calculation formula is as follows:

[0216]

[0217] Among them, S' is the area of ​​leopard spots in the region, S is the fundus area of ​​the corresponding region, and ρ is a dimensionless value, which indicates the proportion of leopard spots in the region.

[0218] Therefore, the system can realize automatic analysis, quantitative evaluation and auxiliary diagnosis of fundus photographs, effectively breaking through the limitations of qualitative analysis of traditional fundus examinations. Through the human-computer collaborative operation system, the system can also provide ophthalmologists with lesion annotation and auxiliary diagnosis and treatment plans, facilitating the accurate and comprehensive prevention and control of fundus diseases. The automatic analysis and evaluation technology realizes fully automatic processing and analysis of fundus photographs, provides detailed lesion information, and assists doctors in scientific diagnosis and treatment. The human-computer collaborative operation system combines artificial intelligence technology and doctors' professional knowledge to provide lesion annotation and auxiliary diagnosis and treatment plans, improve the accuracy and efficiency of diagnosis, and promote the prevention and control of fundus diseases towards precision and comprehensiveness.

[0219] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0220] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for segmenting and quantitatively analyzing fundus image lesions based on data enhancement, characterized in that: The steps include: S1. Image preprocessing: Collect user's fundus photos based on retinal camera; Perform ROI extraction, denoising, normalization and enhancement on the user's fundus photos; Determine a preprocessed user fundus image; S2, optic disc and macula positioning: Localize the optic disc and macula based on deep learning object detection; Train the optic disc and macula detection models; Output the target detection frame based on the optic disc and macula detection model; The center point of the detection frame is used as the center point of the optic disc and macula respectively; S3, Video disc segmentation: The center point of the optic disc obtained by optic disc positioning is taken as the origin; Based on the computer vision attention mechanism, the user's fundus image is transformed into polar coordinates; On the polar coordinate image, the edge detection operator is used to obtain the edge of the optic disc in polar coordinates; Perform polar coordinate inverse transformation on the user's fundus image to obtain the optic disc edge in the rectangular coordinate system and perform fine segmentation on the optic disc; S4. Identification of changes in myopic fundus features: The deep learning semantic segmentation network TransUnet was used to establish the recognition models of leopard spot area, patchy atrophy area and diffuse atrophy area respectively; The samples of the training set were input into the deep learning semantic segmentation network model to segment the leopard spot area, patchy atrophy area and diffuse atrophy area; The test set is used to evaluate the segmentation accuracy of the deep learning semantic segmentation network model, and the accuracy, sensitivity, and specificity are calculated respectively; S5. Quantification of leopard spot area features: Based on the results of ROI and optic disc segmentation, the density of leopard spots within 4.5 mm with the optic disc center as the origin, the density of leopard spots within 1 mm, 3 mm, and 6 mm with the macula center as the origin, and the density of leopard spots in the upper and lower nasal and temporal areas and the upper and lower nasal, lower nasal, upper and lower temporal areas of the corresponding areas were calculated. The calculation formula is as follows: Among them, S' is the area of ​​leopard spots in the region, S is the fundus area of ​​the corresponding region, and ρ is a dimensionless value, which indicates the proportion of leopard spots in the region.

2. The method for segmenting and quantitatively analyzing fundus images based on data enhancement according to claim 1, characterized in that: In S1, the user's fundus photo is subjected to ROI extraction, denoising, normalization and enhancement processing, and the following operations are performed: Obtain user's fundus photo; Based on image cropping and magnification technology, the region of interest (ROI) is extracted from the user's fundus photo; Extract the area containing important information in the user's fundus photo, remove the irrelevant background in the user's fundus photo, and focus the analysis on the key parts of the user's fundus photo; Based on advanced image denoising algorithms, denoise the user's fundus photos; Replace the value of a point in the user's fundus photo with the median value of all points in a region of the point, so that pixels with relatively large grayscale differences from surrounding pixels are replaced with values ​​close to the surrounding pixel values, eliminating isolated noise points and removing noise from the user's fundus photo; Based on histogram equalization and gamma correction technology, normalize the user's fundus photos; Eliminate the brightness and contrast differences between the user's fundus photos to determine a consistent user's fundus image; Based on image enhancement technology, enhance the user's fundus image; The user's fundus image is contrast enhanced and sharpened to highlight the detailed features in the user's fundus image and make the lesion area more obvious.

3. The method for fundus image lesion segmentation and quantitative analysis based on data enhancement according to claim 2, characterized in that: Perform contrast enhancement on the user's fundus image, including: Extracting the adjusted pixel grayscale value and the unadjusted pixel grayscale value corresponding to the noise point of the user's fundus image; Extracting pixel grayscale values ​​corresponding to pixel points adjacent to noise points in the user's fundus image; Obtaining a pixel grayscale adjustment degree coefficient corresponding to each noise point by using the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to each noise point in the user's fundus image and the pixel grayscale values ​​corresponding to the pixel points adjacent to the noise point in the user's fundus image; The pixel grayscale adjustment coefficient corresponding to each noise point is obtained by the following formula: Wherein, L represents the pixel grayscale adjustment coefficient corresponding to each noise point; n represents the total number of pixel points adjacent to the noise point in the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the pixel grayscale value after adjustment and the pixel grayscale value before adjustment corresponding to the noise point; H z represents the pixel grayscale median value of n pixels adjacent to the noise point; g represents the adjustment factor, and the adjustment factor is obtained by the following formula: Wherein, g represents the adjustment factor; n represents the total number of pixel points adjacent to the noise point of the user's fundus image; H i represents the pixel grayscale value of the i-th pixel adjacent to the noise point; H q and H h Indicates the adjusted pixel grayscale value and the pixel grayscale value before adjustment corresponding to the noise point; The contrast enhancement process is performed on the user fundus image by utilizing the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image.

4. The method for segmenting and quantitatively analyzing fundus image lesions based on data enhancement according to claim 3, characterized in that: Performing contrast enhancement processing on the user fundus image by using pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image includes: Extracting pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user's fundus image; Extracting pixel points whose difference between the pixel grayscale value contained in the user fundus image and the average value of the adjusted pixel grayscale value corresponding to the noise point is lower than a preset grayscale difference threshold as target pixel points; Obtaining a contrast adjustment coefficient using the pixel grayscale value corresponding to the target pixel point; The contrast adjustment coefficient is obtained by the following formula: Where k represents the contrast adjustment coefficient; m represents the number of target pixels; H mi Represents the pixel gray value corresponding to the i-th target pixel; H sp Represents the average value of the adjusted pixel grayscale values ​​corresponding to all noise points; H sz Indicates the middle value of the adjusted pixel grayscale value corresponding to the noise point; H min Indicates the minimum grayscale value of the target pixel; H sm Indicates the maximum value of the adjusted pixel grayscale value corresponding to the noise point; Obtaining a target contrast using the contrast adjustment coefficient and the pixel grayscale adjustment degree coefficients corresponding to all noise points contained in the user fundus image; The target contrast is obtained by the following formula: Among them, D m represents the target contrast; D represents the original grayscale value of the user's fundus image; a represents the total number of all noise points contained in the user's fundus image; L i Indicates the pixel grayscale adjustment coefficient corresponding to the i-th noise point; L z represents the middle value of the pixel grayscale adjustment coefficient corresponding to n noise points; k represents the contrast adjustment coefficient; The contrast enhancement adjustment is performed on the user fundus image according to the target contrast.

5. The method for fundus image lesion segmentation and quantitative analysis based on data enhancement according to claim 2, characterized in that: In S2, the optic disc and macula detection models are trained, and the following operations are performed: According to the user's requirements for locating the optic disc and macula in the fundus image, annotated samples of the optic disc and macula are collected, and annotated optic disc samples and macula samples are determined; The network model of single-step detection SSD of target detection is selected, in which the backbone network in the network structure uses ResNet-50 network; The network model of single-step detection SSD of selected target detection is trained based on the labeled optic disc samples and macula samples, and the network model parameters are continuously optimized; The optic disc and macula detection models were determined.

6. The method for fundus image lesion segmentation and quantitative analysis based on data enhancement according to claim 5, characterized in that: In S2, the target detection frame is output based on the optic disc and macula detection model, and the following operations are performed: Obtain the optic disc and macula detection model; Input the preprocessed user fundus image into the optic disc and macula detection model; Perform target detection on the pre-processed user fundus image based on the optic disc and macula detection model, so that the target area can be accurately detected at different scales; Based on the optic disc and macula detection model, the target detection frame is output, and the center point of each area is extracted as the center point of the optic disc and macula.

7. The method for segmenting and quantitatively analyzing fundus image lesions based on data enhancement according to claim 6, characterized in that: In S3, the optic disc is finely segmented by performing the following operations: Polar coordinate conversion: Taking the center of the optic disc as the origin, the user's fundus image is converted into polar coordinates; Edge detection: Use edge detection operators to identify the edges of the optic disc; Edge delineation: The identified edges are converted back into Cartesian coordinates for delineation, and the optic disc area is accurately segmented by delineating the edges.

8. The method for segmenting and quantitatively analyzing fundus image lesions based on data enhancement according to claim 7, characterized in that: In S4, the leopard spot area, patchy atrophy area and diffuse atrophy area are segmented, and the following operations are performed: According to the recognition requirements of myopic fundus features, the samples were labeled by two doctors; One of the physicians provided the initial annotations, and the other physician reviewed and revised them; Among them, the labeled samples include leopard spot areas, patchy atrophy areas, and diffuse atrophy areas; The labeled samples are divided into training set and test set in a ratio of 7:3; During training, the samples of the training set are input into the deep learning semantic segmentation network model; Among them, the deep learning semantic segmentation network model uses the Transformer structure in the encoder to extract high-level features from the input user's fundus image; In the decoding stage, the extracted high-level features are restored to the resolution of the input image by upsampling; The deep learning semantic segmentation network model outputs a confidence probability map corresponding to the extracted high-level features; Threshold segmentation is performed based on the confidence probability map to segment the leopard spot area, patchy atrophy area and diffuse atrophy area.

9. The method for fundus image lesion segmentation and quantitative analysis based on data enhancement according to claim 8, characterized in that: In S4, the accuracy, sensitivity and specificity are calculated by performing the following operations: Taking image pixels as the smallest analysis unit, the accuracy, sensitivity and specificity of leopard spot area, patchy atrophy area and diffuse atrophy area were calculated respectively; Among them, the accuracy, sensitivity, and specificity of the leopard spot region were 0.9766, 0.9551, and 0.9801, respectively; Among them, the accuracy, sensitivity, and specificity of patchy atrophy area were 0.9993, 0.9795, and 0.9995, respectively; Among them, the accuracy, sensitivity and specificity of diffuse atrophy area were 0.997, 0.978 and 0.998 respectively.

10. The method for segmenting and quantitatively analyzing fundus image lesions based on data enhancement according to claim 9, characterized in that: In S5, after the leopard spot density is calculated and quantified, a user fundus quantitative report is generated, wherein the user fundus quantitative report includes detailed information on the area, shape, and severity of the lesion area, and based on the user fundus quantitative report, the doctor is assisted in formulating a personalized treatment plan for the user.

Images