CN-119784773-B - Method and system for partitioning hardened exudates of diabetic retinopathy

CN119784773BCN 119784773 BCN119784773 BCN 119784773BCN-119784773-B

Abstract

The application provides a method and a system for dividing hardened exudates of diabetic retinopathy, which relate to the technical field of image division, and are used for determining the image complexity of a color fundus image, dividing the color fundus image into a plurality of exudation areas of the hard exudates according to the distribution characteristics of the hard exudates in diabetic retinas, determining the edge contrast of each exudation area, determining the pathological change heterogeneity index after diabetic retinopathy through all the edge contrast and the image complexity, determining the convolution fusion boundary of all the exudation areas through the pathological change heterogeneity index and the boundary characteristics of each exudation area, and carrying out convolution fusion division on all the exudation areas of the hard exudates based on the convolution fusion boundary to obtain a plurality of division areas of the hard exudates in the color fundus image.

Inventors

ZHOU JIE
MEI QIANYU
XU BAISHENG
WANG SHANHONG
ZHAO PING

Assignees

浙江省中医药研究院

Dates

Publication Date: 20260505
Application Date: 20241213

Claims (8)

1. A method of partitioning a hardened exudate of diabetic retinopathy, said partitioning method comprising the steps of: collecting a color fundus image after diabetic retinopathy, and further determining the image complexity of the color fundus image; segmenting the color fundus image into a plurality of exudation regions of hard exudates based on the distribution characteristics of hard exudates in the diabetic retina; Determining the edge contrast of each exudation area by combining the attention mechanism of the convolutional neural network with the gray gradient between the hard exudates and the background area in the color fundus image, and determining the lesion heterogeneity index after diabetic retinopathy through all the edge contrast and the image complexity; determining convolution fusion boundaries of all exudation areas according to the lesion heterogeneity index and boundary characteristics of each exudation area, and carrying out convolution fusion segmentation on all exudation areas of the hard exudates based on the convolution fusion boundaries to obtain a plurality of segmentation areas of the hard exudates in the color fundus image; determining the edge contrast of each exudation area based on the attention mechanism of the convolutional neural network in combination with the gray gradient between the hard exudates and the background area in the color fundus image specifically comprises: determining a gray scale gradient between hard exudates and a background region in the color fundus image; For each exudation area, acquiring a gray level difference value between the exudation area and a background area from the gray level gradient; Determining an influence weight of the gray scale difference in the exudation area based on the attention mechanism of the convolutional neural network; Determining the edge contrast of the exudation area according to the influence weight and the gray level difference value, and further determining the edge contrast of each exudation area; the convolution fusion boundary for determining all exudation areas through the lesion heterogeneity index and the boundary characteristics of each exudation area specifically comprises the following steps: for each exudation area, obtaining the boundary characteristics of the exudation area, and further determining the fusion weight of the exudation area; Determining boundary fusion values of the exudation areas according to the fusion weights and the boundary characteristics, and further obtaining the boundary fusion values of the exudation areas; And carrying out convolution operation on all the boundary fusion values to obtain convolution fusion boundaries of all the exudation areas.
2. A method of segmentation of diabetic retinopathy hardening exudates as claimed in claim 1, characterized in that determining the image complexity of the colour fundus image comprises in particular: Performing gray level conversion on the color fundus image to obtain a gray level image; Converting the gray level image into a gray level co-occurrence matrix; and determining the image complexity of the color fundus image through the gray level co-occurrence matrix.
3. A method of segmentation of diabetic retinopathy stiffening exudates as claimed in claim 1, wherein the segmentation of the color fundus image into a plurality of exudation areas of the hard exudates according to the distribution characteristics of the hard exudates in the diabetic retina comprises in particular: Identifying all exudation positions of the color fundus image based on a data labeling mechanism; Screening all hard exudate locations from all exudate locations based on the distribution characteristics of hard exudates in the diabetic retina; multiple exudation areas of hard exudates are determined based on all hard exudate locations.
4. A method of segmentation of diabetic retinopathy stiffening exudates as claimed in claim 1, characterized in that determining the post-diabetic retinopathy lesion heterogeneity index from all edge contrast and said image complexity comprises in particular: initializing a lesion statistical model based on a support vector machine for each exudation area; determining a statistical function of a lesion statistical model according to the image complexity and the edge contrast of the exudation area; carrying out fusion statistics on exudation areas after diabetic retinopathy by using a pathological change statistical model to obtain pathological change heterogeneity of the exudation areas, and further obtaining pathological change heterogeneity of each exudation area; The lesion heterogeneity index after diabetic retinopathy was determined by all lesion heterogeneity.
5. A method of partitioning a diabetic retinopathy stiffening exudate as in claim 1, wherein said stiffening exudate is a spotted material formed by fat and protein deposition within the retina.
6. A method of segmentation of diabetic retinopathy hardened exudates as claimed in claim 1 wherein convolutionally fusing all exudation areas of the hard exudates based on the convolutionally fused interfaces, the obtaining of the plurality of segmented areas of hard exudates in the color fundus image comprises: Determining convolution kernels of various scales based on the shape characteristics of the hard exudate; determining the fusion weights of convolution kernels of various scales through the convolution fusion boundary; And fusing all exudation areas into a plurality of segmented areas of hard exudates in the color fundus image according to all the fusion weights.
7. A method of segmenting diabetic retinopathy stiffening exudate as claimed in claim 1, wherein the high resolution fundus camera is used to capture a colour fundus image after diabetic retinopathy.
8. A diabetic retinopathy stiffening exudate separation system for performing a diabetic retinopathy stiffening exudate separation method as set forth in any one of claims 1 to 7, wherein the separation system comprises: the image acquisition module is used for acquiring a color fundus image after diabetic retinopathy, so as to determine the image complexity of the color fundus image; an initial segmentation module for segmenting the color fundus image into a plurality of exudation regions of hard exudates based on the distribution characteristics of hard exudates in the diabetic retina; the heterogeneity evaluation module is used for determining the edge contrast of each exudation area based on the attention mechanism of the convolutional neural network and combining the gray gradient between the hard exudates and the background area in the color fundus image, and determining the pathological change heterogeneity index after diabetic retinopathy through all the edge contrast and the image complexity; The fusion segmentation module is used for determining convolution fusion boundaries of all exudation areas according to the lesion heterogeneity index and boundary characteristics of each exudation area, and carrying out convolution fusion segmentation on all exudation areas of the hard exudates based on the convolution fusion boundaries to obtain a plurality of segmentation areas of the hard exudates in the color fundus image.

Description

Method and system for partitioning hardened exudates of diabetic retinopathy Technical Field The application relates to the technical field of image segmentation, in particular to a method and a system for segmenting hardened exudates of diabetic retinopathy. Background Diabetic retinopathy is an eye disease caused by diabetes, hard exudates are one of main manifestations of DR, and are formed by fat and protein deposition in retina, so that timely and accurate detection and segmentation of hard exudates are important for disease assessment and treatment scheme establishment, in recent years, with rapid development of medical imaging technology and deep learning algorithm, image segmentation-based hard exudates detection methods are widely studied, and image segmentation technology is helpful for doctors to rapidly identify and diagnose diseases by accurately segmenting different areas in medical images and extracting lesion area information. Conventional convolutional neural networks have limitations in extracting complex edge and detail information, especially in the segmentation of hard exudates, fine visual features are easily ignored. The convolution fusion method enables a network to synthesize information of different layers through combination of multiple layers of features, key features are extracted more accurately, meanwhile, a diabetic retina image often contains noise or a complex background and can interfere detection of hard exudates, and convolution fusion segmentation is beneficial to enhancing separation of signals and noise through multi-channel feature fusion, stability of segmentation results in complex scenes is improved, so that the difficulty in industry is how to achieve convolution fusion segmentation of diabetic retinopathy hard exudates, and accordingly the segmentation accuracy of the diabetic retinopathy hard exudates is improved. Disclosure of Invention The application provides a method and a system for dividing hardened exudates of diabetic retinopathy, which can realize convolution fusion division of the hardened exudates of diabetic retinopathy, thereby improving the division precision of the hardened exudates after diabetic retinopathy. In a first aspect, the present application provides a method of partitioning a diabetic retinopathy stiffening exudate, the partitioning method comprising the steps of: collecting a color fundus image after diabetic retinopathy, and further determining the image complexity of the color fundus image; segmenting the color fundus image into a plurality of exudation regions of hard exudates based on the distribution characteristics of hard exudates in the diabetic retina; Determining the edge contrast of each exudation area by combining the attention mechanism of the convolutional neural network with the gray gradient between the hard exudates and the background area in the color fundus image, and determining the lesion heterogeneity index after diabetic retinopathy through all the edge contrast and the image complexity; And determining convolution fusion boundaries of all exudation areas according to the lesion heterogeneity index and boundary characteristics of each exudation area, and carrying out convolution fusion segmentation on all exudation areas of the hard exudates based on the convolution fusion boundaries to obtain a plurality of segmentation areas of the hard exudates in the color fundus image. In this embodiment, determining the image complexity of the color fundus image specifically includes: Performing gray level conversion on the color fundus image to obtain a gray level image; Converting the gray level image into a gray level co-occurrence matrix; and determining the image complexity of the color fundus image through the gray level co-occurrence matrix. In this embodiment, the dividing the color fundus image into a plurality of exudation areas of hard exudates based on the distribution characteristics of hard exudates in the diabetic retina specifically includes: Identifying all exudation positions of the color fundus image based on a data labeling mechanism; Screening all hard exudate locations from all exudate locations based on the distribution characteristics of hard exudates in the diabetic retina; multiple exudation areas of hard exudates are determined based on all hard exudate locations. In this embodiment, determining the edge contrast of each exudation area based on the attention mechanism of the convolutional neural network in combination with the gray scale gradient between the hard exudates and the background area in the color fundus image specifically includes: determining a gray scale gradient between hard exudates and a background region in the color fundus image; For each exudation area, acquiring a gray level difference value between the exudation area and a background area from the gray level gradient; Determining an influence weight of the gray scale difference in the exudation area based on the attention m