Diabetic retinopathy hard exudate segmentation method and system

Abstract

This application provides a method and system for segmenting hard exudates in diabetic retinopathy, relating to the field of image segmentation technology. The method involves determining the image complexity of a color fundus image; segmenting the color fundus image into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in the diabetic retina; determining the edge contrast of each exudation region; determining the variability index of diabetic retinopathy based on all edge contrasts and image complexity; determining the convolution fusion boundary of all exudation regions based on the variability index and the boundary characteristics of each exudation region; and performing convolution fusion segmentation on all exudation regions of hard exudates based on the convolution fusion boundary to obtain multiple segmented regions of hard exudates in the color fundus image. This application can achieve convolution fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates in diabetic retinopathy.

Description

Technical Field

[0001] This application relates to the field of image segmentation technology, and more specifically, to a method and system for segmenting hard exudates in diabetic retinopathy. Background Technology

[0002] Diabetic retinopathy is an eye disease caused by diabetes. Hard exudates are one of the main manifestations of DR, formed by the deposition of fat and protein in the retina, marking the progression of the disease. Therefore, timely and accurate detection and segmentation of hard exudates are crucial for disease assessment and treatment planning. In recent years, with the rapid development of medical imaging technology and deep learning algorithms, image segmentation-based methods for detecting hard exudates have been extensively studied. Image segmentation technology can help doctors quickly identify and diagnose diseases by accurately segmenting different regions in medical images and extracting information about lesion areas.

[0003] Traditional convolutional neural networks have limitations in extracting complex edge and detail information, especially in the segmentation of hard exudates, where subtle visual features are easily overlooked. Convolutional fusion methods, by combining features from multiple layers, enable the network to integrate information from different levels and extract key features more accurately. Furthermore, diabetic retinal images often contain noise or complex backgrounds, which can interfere with the detection of hard exudates. Convolutional fusion segmentation, through multi-channel feature fusion, helps enhance the separation of signal and noise, improving the stability of segmentation results in complex scenes. Therefore, how to achieve convolutional fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates after diabetic retinopathy, is a challenge facing the industry. Summary of the Invention

[0004] This application provides a method and system for segmenting hard exudates in diabetic retinopathy, which can achieve convolutional fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates after diabetic retinopathy.

[0005] In a first aspect, this application provides a method for segmenting hard exudates in diabetic retinopathy, the segmentation method comprising the following steps:

[0006] Acquire color fundus images after diabetic retinopathy, and then determine the image complexity of the color fundus images;

[0007] Based on the distribution characteristics of hard exudates in diabetic retinopathy, the color fundus image is segmented into multiple exudation regions of hard exudates;

[0008] The attention mechanism based on convolutional neural networks is combined with the gray-level gradient between hard exudates and background areas in the color fundus image to determine the edge contrast of each exudate area. The pathological qualitative indicators of diabetic retinopathy are determined by all edge contrasts and the image complexity.

[0009] The convolutional fusion boundaries of all exudative regions are determined by the morphological indicators of the pathogenesis and the boundary characteristics of each exudative region. Based on the convolutional fusion boundaries, all exudative regions of hard exudate are segmented by convolutional fusion to obtain multiple segmented regions of hard exudate in the color fundus image.

[0010] In this embodiment, determining the image complexity of the color fundus image specifically includes:

[0011] The color fundus image is converted to grayscale to obtain a grayscale image;

[0012] Convert the grayscale image into a grayscale co-occurrence matrix;

[0013] The image complexity of the color fundus image is determined by the gray-level co-occurrence matrix.

[0014] In this embodiment, segmenting the color fundus image into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in diabetic retinopathy specifically includes:

[0015] All exudation locations in the color fundus image were identified based on a data annotation mechanism;

[0016] Based on the distribution characteristics of hard exudates in diabetic retinopathy, all hard exudate locations were screened from all exudate locations;

[0017] Multiple exudation zones of hard exudates were determined based on the locations of all hard exudates.

[0018] In this embodiment, the edge contrast of each exudate region is determined by the attention mechanism based on the convolutional neural network, combined with the gray-level gradient between the hard exudate and the background region in the color fundus image. Specifically, this includes:

[0019] Determine the grayscale gradient between the hard exudate and the background area in the color fundus image;

[0020] For each infiltration region, the grayscale difference value between the infiltration region and the background region is obtained from the grayscale gradient;

[0021] The influence weight of grayscale differences in the exudation region is determined based on the attention mechanism of convolutional neural networks;

[0022] The edge contrast of the exudation area is determined by the influence weight and the grayscale difference value, and then the edge contrast of each exudation area is determined.

[0023] In this embodiment, determining the pathological qualitative indicators of diabetic retinopathy based on all edge contrasts and the image complexity specifically includes:

[0024] For each exudation zone, initialize a lesion statistical model based on support vector machine;

[0025] The statistical function of the lesion statistical model is determined based on the image complexity and the edge contrast of the exudation area.

[0026] Using this pathological statistical model, we performed fusion statistics on the exudative areas after diabetic retinopathy to obtain the pathological variability of the exudative areas, and then obtained the pathological variability of each exudative area.

[0027] Determine the pathogenicity markers for diabetic retinopathy by examining all pathogenicity characteristics.

[0028] In this embodiment, the hard exudate is a speckled substance formed by the deposition of fat and protein in the retina.

[0029] In this embodiment, determining the convolutional fusion boundary of all exudation regions using the aforementioned variability qualitative indicators and the boundary characteristics of each exudation region specifically includes:

[0030] For each infiltration region, the boundary characteristics of the infiltration region are obtained, and then the fusion weight of the infiltration region is determined.

[0031] The boundary fusion value of the infiltration region is determined by the fusion weight and the boundary characteristics, thereby obtaining the boundary fusion value of each infiltration region;

[0032] Perform a convolution operation on all boundary fusion values ​​to obtain the convolution fusion boundaries of all exudation regions.

[0033] In this embodiment, the convolutional fusion segmentation of all exudate regions of hard exudate based on the convolutional fusion boundary to obtain multiple segmented regions of hard exudate in the color fundus image specifically includes:

[0034] Convolution kernels of various scales were determined based on the shape characteristics of the hard exudate;

[0035] The fusion weights of convolution kernels at various scales are determined by the convolution fusion bound.

[0036] Based on all fusion weights, all exudate regions are fused into multiple segmented regions of hard exudate in the color fundus image.

[0037] In this embodiment, a high-resolution fundus camera is used to acquire color fundus images after diabetic retinopathy.

[0038] Secondly, this application provides a segmentation system for hard exudates in diabetic retinopathy, used to perform a method for segmenting hard exudates in diabetic retinopathy, the segmentation system comprising:

[0039] An image acquisition module is used to acquire color fundus images after diabetic retinopathy, and then determine the image complexity of the color fundus images;

[0040] The initial segmentation module is used to segment the color fundus image into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in diabetic retinopathy;

[0041] The heterogeneity assessment module is used to determine the edge contrast of each exudate region by combining the gray-level gradient between the hard exudate and the background region in the color fundus image based on the attention mechanism of the convolutional neural network, and to determine the pathogenesis heterogeneity index after diabetic retinopathy by using all edge contrasts and the image complexity.

[0042] The fusion segmentation module is used to determine the convolutional fusion boundary of all exudate regions through the morphological indicators of the pathogenesis and the boundary characteristics of each exudate region, and to perform convolutional fusion segmentation on all exudate regions of hard exudate based on the convolutional fusion boundary to obtain multiple segmented regions of hard exudate in the color fundus image.

[0043] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0044] By acquiring color fundus images after diabetic retinopathy, the image complexity of the color fundus images is determined. Based on the distribution characteristics of hard exudates in the diabetic retina, the color fundus images are segmented into multiple exudation regions of hard exudates. Based on the attention mechanism of a convolutional neural network and the gray-level gradient between the hard exudates and the background region in the color fundus images, the edge contrast of each exudation region is determined. The pathological variability index after diabetic retinopathy is determined by all edge contrasts and the image complexity. The convolutional fusion boundary of all exudation regions is determined by the pathological variability index and the boundary characteristics of each exudation region. Based on the convolutional fusion boundary, all exudation regions of hard exudates are segmented by convolutional fusion to obtain multiple segmented regions of hard exudates in the color fundus images.

[0045] Therefore, in this application, the convolutional fusion boundaries of all exudate regions are determined by the qualitative indicators of pathological changes and the boundary characteristics of each exudate region. Based on these boundaries, convolutional fusion segmentation is performed on all exudate regions of hard exudate to obtain multiple segmented regions of hard exudate in the color fundus image. First, data annotation and feature-based filtering can accurately extract regions with hard exudate characteristics, avoiding unnecessary noise interference. Furthermore, combining image complexity with the boundary characteristics of the exudate region makes the segmented regions more targeted and reliable, providing accurate input data for subsequent convolutional fusion segmentation. This allows the convolutional network to focus more on relevant regions during processing, helping to reduce... Misclassification and omission are addressed to ensure that the convolutional fusion segmentation algorithm can perform refined analysis of hard exudates, thereby improving segmentation accuracy and detection effectiveness. Then, by utilizing the attention mechanism and gray-level gradient analysis based on convolutional neural networks, the edge characteristics of different exudate regions can be evaluated, revealing the differences in edge characteristics in color fundus images. Pathogenicity indicators reflect the variation characteristics of hard exudates in different regions, providing weight and regional characteristic basis for convolutional fusion operations. When performing convolutional fusion segmentation in conjunction with pathogenicity indicators, it is helpful to dynamically adjust and refine the fusion according to the actual complexity of the lesion region, optimize the accuracy of the fusion results, and thus improve the accuracy of hard exudate segmentation after diabetic retinopathy.

[0046] In summary, the technical solution adopted in this application can achieve convolutional fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates after diabetic retinopathy. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the method for segmenting hard exudates in diabetic retinopathy provided in this application;

[0049] Figure 2 This is an exemplary flowchart for determining edge contrast provided in this application;

[0050] Figure 3 This is an exemplary flowchart for determining convolutional fusion boundaries provided in this application;

[0051] Figure 4 This is a modular structure diagram of the hard exudate segmentation system for diabetic retinopathy provided in this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] This application provides a method and system for segmenting hard exudates in diabetic retinopathy. The core of the method involves acquiring color fundus images of diabetic retinopathy and determining the image complexity of these images. Based on the distribution characteristics of hard exudates in the diabetic retina, the color fundus image is segmented into multiple exudation regions of hard exudates. Using an attention mechanism based on a convolutional neural network and the gray-level gradient between the hard exudates and the background region in the color fundus image, the edge contrast of each exudation region is determined. The pathological variability index of diabetic retinopathy is determined using all edge contrasts and the image complexity. The convolutional fusion boundary of all exudation regions is determined using the pathological variability index and the boundary characteristics of each exudation region. Based on the convolutional fusion boundary, all exudation regions of the hard exudate are segmented using convolutional fusion to obtain multiple segmented regions of hard exudates in the color fundus image. This method can achieve convolutional fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates in diabetic retinopathy.

[0054] Example 1

[0055] To better understand the above technical solutions, a detailed description of the technical solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a method for segmenting hard exudates in diabetic retinopathy according to this embodiment of the present application. The segmentation method includes the following steps:

[0056] In step S1, a color fundus image of the eye following diabetic retinopathy is acquired, and the image complexity of the color fundus image is then determined.

[0057] It should be noted that color fundus images are retinal images used to examine the structure and health of the eye; in practice, high-resolution fundus cameras can be used to acquire color fundus images after diabetic retinopathy.

[0058] In this embodiment, the image complexity of the color fundus image can be determined in the following way:

[0059] The color fundus image is converted to grayscale to obtain a grayscale image;

[0060] Convert the grayscale image into a grayscale co-occurrence matrix;

[0061] The image complexity of the color fundus image is determined by the gray-level co-occurrence matrix.

[0062] It should be noted that image complexity is an indicator of the richness of image detail and texture information in a color fundus image; the gray-level co-occurrence matrix represents the matrix of spatial relationships between pixel gray levels in a color fundus image; and a grayscale image is a single-channel image of brightness values ​​in a color fundus image.

[0063] In practice, firstly, an image processing library (such as cv2.cvtColor() in OpenCV) can be used to perform grayscale conversion on the color fundus image, and the result of the grayscale conversion is used as a grayscale image. Then, the co-occurrence frequency of pixel gray levels in the grayscale image at specific directions and distances is statistically analyzed. Fixed offset parameters (such as horizontal, vertical, or diagonal directions) are set based on historical experience, and a matrix is ​​generated using statistical methods and normalized to ensure that the sum of the elements of the matrix is ​​1. This matrix can be used as the grayscale co-occurrence matrix. Finally, the information entropy feature in the grayscale co-occurrence matrix can be used as the image complexity of the color fundus image. The information entropy feature reflects the texture and structural characteristics of the image. High complexity means that the image has rich details and information, while low complexity indicates that the texture of the image is relatively simple.

[0064] In step S2, the color fundus image is segmented into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in diabetic retinopathy.

[0065] In this embodiment, the color fundus image is segmented into multiple exudate regions of hard exudate based on the distribution characteristics of hard exudate in diabetic retinopathy, which can be achieved in the following manner:

[0066] All exudation locations in the color fundus image were identified based on a data annotation mechanism;

[0067] Based on the distribution characteristics of hard exudates in diabetic retinopathy, all hard exudate locations were screened from all exudate locations;

[0068] Multiple exudation zones of hard exudates were determined based on the locations of all hard exudates.

[0069] It should be noted that in this application, the exudation region is an independent region formed by the aggregation of multiple hard exudates, which are spot-like substances formed by the deposition of fat and protein in the retina. In specific implementation, firstly, a labeling tool (e.g., Labelbox) is used for semi-automatic labeling to mark the exudation regions in the color fundus image. In other embodiments, expert experience and pre-trained deep learning models can be combined to improve the accuracy of labeling, which is not limited here. Then, hard exudates after diabetic retinopathy usually have specific colors, shapes and distribution patterns, such as white or yellow spots or patches, and are usually concentrated in specific areas of the retina. Therefore, the set of specific colors, shapes and distribution patterns can be used as the distribution features of hard exudates in diabetic retina in the target detection algorithm, so that the target detection algorithm can be used to filter out all hard exudate locations from all exudate locations. Finally, a region growing algorithm can be used to aggregate adjacent hard exudate locations into complete exudation regions, thus obtaining multiple exudation regions of hard exudates.

[0070] In step S3, the edge contrast of each exudate region is determined by combining the gray-level gradient between the hard exudate and the background region in the color fundus image based on the attention mechanism of the convolutional neural network. The pathological qualitative indicators of diabetic retinopathy are determined by all edge contrasts and the image complexity.

[0071] Preferably, in this embodiment, reference Figure 2 As shown, this figure is an exemplary flowchart for determining edge contrast in an embodiment of this application. In this embodiment, the edge contrast of each exudate region is determined based on the attention mechanism of a convolutional neural network combined with the gray-level gradient between the hard exudate and the background region in the color fundus image. The specific steps for determining the edge contrast of each exudate region can be as follows:

[0072] First, in step S31, the grayscale gradient between the hard exudate and the background area in the color fundus image is determined;

[0073] Then, in step S32, for each exudation area, the gray level difference value between the exudation area and the background area is obtained from the gray level gradient;

[0074] Furthermore, in step S33, the influence weight of grayscale differences in the exudation region is determined based on the attention mechanism of the convolutional neural network;

[0075] Finally, in step S34, the edge contrast of the oozing region is determined by the influence weight and the grayscale difference value, thereby determining the edge contrast of each oozing region.

[0076] It should be noted that, in this application, edge contrast represents the degree of grayscale difference between two exudation regions in a color fundus image; grayscale gradient represents the degree of spatial variation of the grayscale value of each pixel in a color fundus image; grayscale difference value is a quantitative value used to measure the contrast between the exudation region and the background region; influence weight reflects the importance of different exudation regions in the convolutional fusion of the color fundus image.

[0077] In practice, firstly, by performing grayscale conversion on the color fundus image and applying an edge detection algorithm (e.g., the Sobel operator), all possible values ​​of the grayscale gradient between the hard exudate and the background region can be determined. Secondly, for each exudate region, the grayscale gradient between the hard exudate and the background region in the exudate region is obtained from the grayscale gradient as the grayscale difference value between the exudate region and the background region. Then, based on the attention mechanism in the convolutional neural network, the grayscale differences in the exudate region are weighted to determine the influence weight of the grayscale differences in the exudate region. This attention mechanism enables the model to automatically adjust the weights, highlight important regions in the image, and optimize the analysis results. Finally, the product of the influence weight and the grayscale difference value can be used as the edge contrast of the exudate region. The edge contrast of each exudate region can be obtained in the above way.

[0078] In this embodiment, the determination of the pathological qualitative indicators of diabetic retinopathy based on all edge contrasts and the image complexity can be achieved in the following manner:

[0079] For each exudation zone, initialize a lesion statistical model based on support vector machine;

[0080] The statistical function of the lesion statistical model is determined based on the image complexity and the edge contrast of the exudation area.

[0081] Using this pathological statistical model, we performed fusion statistics on the exudative areas after diabetic retinopathy to obtain the pathological variability of the exudative areas, and then obtained the pathological variability of each exudative area.

[0082] Determine the pathogenicity markers for diabetic retinopathy by examining all pathogenicity characteristics.

[0083] It should be noted that in this application, the variability index is a feature value that quantifies the overall severity of the lesion. Specifically, firstly, for each exudative region, a lesion statistical model can be initialized using a support vector machine (SVM). This lesion statistical model is a deep learning model used to evaluate the lesion features in the exudative region. This model comprehensively considers the features within the exudative region and uses a SVM classifier or regressor to generate a fused statistical value. Secondly, the statistical function of this lesion statistical model is trained using image complexity and the edge contrast of the exudative region. This statistical function represents the relationship between different pixels and the lesion features within the exudative region. Then, this lesion statistical model is used to perform fused statistics on the exudative regions after diabetic retinopathy. The quantified value of the fused statistics can be used as the variability of the exudative region. The variability of each exudative region can be obtained in the above way. The variability represents the degree and complexity of change in different parts of the lesion region. Finally, the standard deviation of all variability can be used as the variability index after diabetic retinopathy.

[0084] In step S4, the convolutional fusion boundary of all exudate regions is determined by the morphological index of the lesion and the boundary characteristics of each exudate region. Based on the convolutional fusion boundary, all exudate regions of hard exudate are segmented by convolutional fusion to obtain multiple segmented regions of hard exudate in the color fundus image.

[0085] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the convolutional fusion boundary in an embodiment of this application. In this embodiment, determining the convolutional fusion boundary of all exudation regions through the pathogenic qualitative index and the boundary characteristics of each exudation region can be achieved by the following steps:

[0086] First, in step S41, for each infiltration region, the boundary characteristics of the infiltration region are obtained, and then the fusion weight of the infiltration region is determined;

[0087] Then, in step S42, the boundary fusion value of the infiltration region is determined by the fusion weight and the boundary characteristics, thereby obtaining the boundary fusion value of each infiltration region;

[0088] Finally, in step S43, a convolution operation is performed on all boundary fusion values ​​to obtain the convolution fusion boundaries of all exudation regions.

[0089] It should be noted that in this application, the convolution fusion boundary is the boundary that reflects the overall characteristics of the lesion area; the fusion weight is a weight value used to represent the importance of boundary characteristics in the fusion calculation; the boundary characteristics represent the geometric texture characteristics of the edge of the exudation area; and the boundary fusion value refers to the characteristic value of the boundary of the exudation area.

[0090] In specific implementation, firstly, for each exudation region, an edge detection algorithm (such as the Canny operator) can be used to identify the boundary information of hard exudates in the color fundus image as the boundary characteristics of the exudation region. The fusion weights of the exudation regions can be preset based on historical experience. Then, an edge fusion model is initialized, with the fusion weights used as weight parameters in the edge fusion model and the boundary characteristics used as fusion features in the edge fusion model. The boundary influence of the exudation region is quantitatively evaluated using this edge fusion model, and the result of the quantitative evaluation can be used as the boundary fusion value of the exudation region. The boundary fusion value of each exudation region can be obtained in the above way. Finally, a standard two-dimensional convolution method can be used to perform convolution operation on all boundary fusion values, and the result of the convolution operation can be used as the convolution fusion boundary of all exudation regions.

[0091] In this embodiment, the convolutional fusion segmentation of all exudate regions of hard exudate based on the convolutional fusion boundary to obtain multiple segmented regions of hard exudate in the color fundus image can be specifically achieved in the following manner:

[0092] Convolution kernels of various scales were determined based on the shape characteristics of the hard exudate;

[0093] The fusion weights of convolution kernels at various scales are determined by the convolution fusion bound.

[0094] Based on all fusion weights, all exudate regions are fused into multiple segmented regions of hard exudate in the color fundus image.

[0095] In specific implementation, firstly, standard convolutional kernels can be trained using the shape features of hard exudates to obtain training results at various scales. All training results are then used as convolutional kernels at corresponding scales to obtain convolutional kernels at multiple scales. Next, the weight distribution within the convolutional fusion boundary is used as the weight calculation rule for edge intensity and region distribution. This calculation rule is used to adjust the weights of convolutional kernels at different scales, and the adjusted results are used as the fusion weights of the convolutional kernels at the corresponding scales, thus obtaining the fusion weights of convolutional kernels at various scales. Finally, a convolutional fusion model is initialized, with each convolutional kernel used as a convolutional kernel in the model, and the fusion weights used as the scale weights in this model. This convolutional fusion model is then used to perform multi-scale fusion on all exudate regions, and the results of multi-scale fusion can be used as all segmented regions, thus obtaining multiple segmented regions of hard exudates in the color fundus image.

[0096] Therefore, in this application, the convolutional fusion boundaries of all exudate regions are determined by the qualitative indicators of pathological changes and the boundary characteristics of each exudate region. Based on these boundaries, convolutional fusion segmentation is performed on all exudate regions of hard exudate to obtain multiple segmented regions of hard exudate in the color fundus image. First, data annotation and feature-based filtering can accurately extract regions with hard exudate characteristics, avoiding unnecessary noise interference. Furthermore, combining image complexity with the boundary characteristics of the exudate region makes the segmented regions more targeted and reliable, providing accurate input data for subsequent convolutional fusion segmentation. This allows the convolutional network to focus more on relevant regions during processing, helping to reduce... Misclassification and omission are addressed to ensure that the convolutional fusion segmentation algorithm can perform refined analysis of hard exudates, thereby improving segmentation accuracy and detection effectiveness. Then, by utilizing the attention mechanism and gray-level gradient analysis based on convolutional neural networks, the edge characteristics of different exudate regions can be evaluated, revealing the differences in edge characteristics in color fundus images. Pathogenicity indicators reflect the variation characteristics of hard exudates in different regions, providing weight and regional characteristic basis for convolutional fusion operations. When performing convolutional fusion segmentation in conjunction with pathogenicity indicators, it is helpful to dynamically adjust and refine the fusion according to the actual complexity of the lesion region, optimize the accuracy of the fusion results, and thus improve the accuracy of hard exudate segmentation after diabetic retinopathy.

[0097] In summary, the technical solution adopted in this application can achieve convolutional fusion segmentation of hard exudates in diabetic retinopathy, thereby improving the segmentation accuracy of hard exudates after diabetic retinopathy.

[0098] Example 2

[0099] This application provides a system for segmenting hard exudates in diabetic retinopathy, referenced... Figure 4 As shown in the figure, this is a schematic diagram of a segmentation system according to this embodiment of the present application. The segmentation system includes:

[0100] The image acquisition module 100 is used to acquire color fundus images after diabetic retinopathy, and then determine the image complexity of the color fundus images.

[0101] The initial segmentation module 200 is used to segment the color fundus image into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in diabetic retinopathy;

[0102] The heterogeneity assessment module 300 is used to determine the edge contrast of each exudate region by combining the gray-level gradient between the hard exudate and the background region in the color fundus image based on the attention mechanism of the convolutional neural network, and to determine the pathogenesis heterogeneity index after diabetic retinopathy by using all edge contrasts and the image complexity.

[0103] The fusion segmentation module 400 is used to determine the convolutional fusion boundary of all exudate regions through the lesion qualitative index and the boundary characteristics of each exudate region, and to perform convolutional fusion segmentation on all exudate regions of hard exudate based on the convolutional fusion boundary to obtain multiple segmented regions of hard exudate in the color fundus image.

[0104] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0105] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0106] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for segmenting hard exudates in diabetic retinopathy, characterized in that, The segmentation method includes the following steps: Acquire color fundus images after diabetic retinopathy, and then determine the image complexity of the color fundus images; Based on the distribution characteristics of hard exudates in diabetic retinopathy, the color fundus image is segmented into multiple exudation regions of hard exudates; The attention mechanism based on convolutional neural networks is combined with the gray-level gradient between hard exudates and background areas in the color fundus image to determine the edge contrast of each exudate area. The pathological qualitative indicators of diabetic retinopathy are determined by all edge contrasts and the image complexity. The convolutional fusion boundaries of all exudates are determined by the morphological indicators of the disease and the boundary characteristics of each exudate region. Based on the convolutional fusion boundaries, all exudate regions of hard exudate are segmented by convolutional fusion to obtain multiple segmented regions of hard exudate in the color fundus image. The attention mechanism based on convolutional neural networks, combined with the gray-level gradient between hard exudates and the background region in the color fundus image, determines the edge contrast of each exudate region, specifically including: Determine the grayscale gradient between the hard exudate and the background area in the color fundus image; For each infiltration region, the grayscale difference value between the infiltration region and the background region is obtained from the grayscale gradient; The influence weight of grayscale differences in the exudation region is determined based on the attention mechanism of convolutional neural networks; The edge contrast of the seepage area is determined by the influence weight and the grayscale difference value, and then the edge contrast of each seepage area is determined. The convolutional fusion boundary of all exudation regions is determined by using the aforementioned variability indicators and the boundary characteristics of each exudation region. Specifically, this includes: For each infiltration region, the boundary characteristics of the infiltration region are obtained, and then the fusion weight of the infiltration region is determined. The boundary fusion value of the infiltration region is determined by the fusion weight and the boundary characteristics, thereby obtaining the boundary fusion value of each infiltration region; Perform a convolution operation on all boundary fusion values ​​to obtain the convolution fusion boundaries of all exudation regions.

2. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, Determining the image complexity of the color fundus image specifically includes: The color fundus image is converted to grayscale to obtain a grayscale image; Convert the grayscale image into a grayscale co-occurrence matrix; The image complexity of the color fundus image is determined by the gray-level co-occurrence matrix.

3. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, Based on the distribution characteristics of hard exudates in diabetic retinopathy, the color fundus image is segmented into multiple exudation regions of hard exudates, specifically including: All exudation locations in the color fundus image were identified based on a data annotation mechanism; Based on the distribution characteristics of hard exudates in diabetic retinopathy, all hard exudate locations were screened from all exudate locations; Multiple exudation zones of hard exudates were determined based on the locations of all hard exudates.

4. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, The specific qualitative indicators of pathological changes following diabetic retinopathy, determined by considering all edge contrast and image complexity, include: For each exudation zone, initialize a lesion statistical model based on support vector machine; The statistical function of the lesion statistical model is determined based on the image complexity and the edge contrast of the exudation area. The pathological statistical model was used to perform fusion statistics on the exudative areas after diabetic retinopathy to obtain the pathological variability of the exudative areas, and then to obtain the pathological variability of each exudative area. Determine the pathogenicity markers for diabetic retinopathy by examining all pathogenicity characteristics.

5. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, The hard exudate is a speckled substance formed by the deposition of fat and protein in the retina.

6. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, Based on the convolutional fusion boundary, all exudate regions of the hard exudate are segmented by convolutional fusion to obtain multiple segmented regions of the hard exudate in the color fundus image, specifically including: Convolution kernels of various scales were determined based on the shape characteristics of the hard exudate; The fusion weights of convolution kernels at various scales are determined by the convolution fusion bound. Based on all fusion weights, all exudate regions are fused into multiple segmented regions of hard exudate in the color fundus image.

7. The method for segmenting hard exudates in diabetic retinopathy as described in claim 1, characterized in that, High-resolution fundus cameras were used to acquire color fundus images of the eye following diabetic retinopathy.

8. A system for segmenting hard exudates in diabetic retinopathy, used to perform a method for segmenting hard exudates in diabetic retinopathy as described in any one of claims 1 to 7, characterized in that, The segmentation system includes: An image acquisition module is used to acquire color fundus images after diabetic retinopathy, and then determine the image complexity of the color fundus images; The initial segmentation module is used to segment the color fundus image into multiple exudation regions of hard exudates based on the distribution characteristics of hard exudates in diabetic retinopathy; The heterogeneity assessment module is used to determine the edge contrast of each exudate region by combining the gray-level gradient between the hard exudate and the background region in the color fundus image based on the attention mechanism of the convolutional neural network, and to determine the pathogenesis heterogeneity index after diabetic retinopathy by using all edge contrasts and the image complexity. The fusion segmentation module is used to determine the convolutional fusion boundary of all exudate regions through the morphological indicators of the pathogenesis and the boundary characteristics of each exudate region, and to perform convolutional fusion segmentation on all exudate regions of hard exudate based on the convolutional fusion boundary to obtain multiple segmented regions of hard exudate in the color fundus image.

Images