Fundus blood vessel intelligent segmentation method and system based on retinal angiography

Abstract

The invention provides an intelligent fundus blood vessel segmentation method and system based on retinal angiography. The method comprises the following steps: acquiring a fundus image through a retinal angiography technology; performing preprocessing operation on the eye fundus image to obtain a preprocessed image; performing super-pixel segmentation on the preprocessed image to obtain segmented regions; performing fundus blood vessel recognition on the segmented region to obtain a recognition result; and constructing a trained fundus blood vessel segmentation model, and performing feature segmentation on the recognition result through the fundus blood vessel segmentation model to obtain a segmentation result. According to the method, multi-level feature extraction and fusion are carried out through the intelligent segmentation model, effective extraction of multi-dimensional features such as color, texture, shape and strength of the eye fundus image is realized, and the segmentation accuracy is improved through feature fusion.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method and system for intelligent segmentation of fundus blood vessels based on retinal angiography. Background Art

[0002] The detection and analysis of fundus blood vessels are of great significance in clinical medicine, especially in the diagnosis of ophthalmology and related diseases such as diabetes and cardiovascular disease. At present, technologies such as retinal angiography (FFA) and optical coherence tomography (OCT) are widely used to obtain fundus vascular images. Retinal angiography technology can provide clear imaging of fundus blood vessels. By analyzing these images, information about arteriovenous vessels and early warning of potential diseases can be obtained. However, due to the complexity of fundus images, there are still many challenges in automating this process.

[0003] First, the existing fundus vascular segmentation methods often rely on manual feature extraction, resulting in insufficient segmentation accuracy, inability to fully capture complex structures, and poor generalization of segmentation results. Secondly, many traditional algorithms have low execution efficiency and cannot meet the needs of real-time processing, especially when processing full-field images. The bottleneck of processing speed significantly affects the feasibility of clinical applications. In addition, fundus images are often affected by noise and artifacts. The existing technology is insufficient in noise removal and artifact processing, which reduces the reliability of vascular segmentation. The existing technology also lacks intelligent analysis capabilities. Many methods only focus on image segmentation itself and do not conduct a comprehensive analysis of vascular features (such as arteriovenous classification and vascular diameter calculation, etc.), which limits its ability to provide more comprehensive physiological and pathological information. Finally, many segmentation models are trained on specific data sets and lack the ability to adapt to different patients and different types of images, resulting in poor performance on new data. Therefore, it is very necessary to design an intelligent fundus vascular segmentation method and system based on retinal angiography. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for intelligent segmentation of fundus blood vessels based on retinal angiography, so as to perform multi-level feature extraction and fusion through an intelligent segmentation model, realize the effective extraction of multi-dimensional features such as color, texture, shape and intensity of fundus images, and improve the accuracy of segmentation through feature fusion.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for intelligent segmentation of fundus blood vessels based on retinal angiography comprises the following steps:

[0007] Obtain fundus images through retinal angiography technology; fundus images include: full-field images and local enhanced images;

[0008] Performing a preprocessing operation on the fundus image to obtain a preprocessed image;

[0009] Perform superpixel segmentation on the preprocessed image to obtain the segmented area;

[0010] Perform fundus blood vessel recognition on the segmented area to obtain recognition results; the recognition results include: optic disc diameter, arteriovenous blood vessel classification results and arteriovenous ratio;

[0011] A trained fundus vascular segmentation model is constructed, and the recognition result is subjected to feature segmentation through the fundus vascular segmentation model to obtain a segmentation result; the training process of the fundus vascular segmentation model includes: feature extraction, feature integration and loss function optimization.

[0012] Optionally, the full-field image is an image covering the entire fundus range; the size of the local enhanced image is 1024*1024 pixels, and the number of blood vessels in the local enhanced image is greater than 20.

[0013] Optionally, performing a preprocessing operation on the fundus image to obtain a preprocessed image includes:

[0014] The fundus image is denoised by using a Gaussian filter to obtain a denoised image;

[0015] Perform histogram equalization on the denoised image to obtain an enhanced image;

[0016] The enhanced image is binarized to obtain a preprocessed image.

[0017] Optionally, superpixel segmentation is performed on the preprocessed image to obtain segmented regions, including:

[0018] Determine the number of superpixels based on the size of the preprocessed image;

[0019] The superpixel size is obtained according to the number of superpixels, and the seed point is determined according to the superpixel size;

[0020] Extract the seed vector of the seed point, and perform superpixel allocation according to the distance between the seed vectors to obtain the allocation result;

[0021] The superpixel is updated according to the allocation result to obtain the segmented area.

[0022] Optionally, the distance between seed vectors is calculated as: Where Dis is the vector distance, (x p ,y p ) is the position of pixel p, is the seed point c k Location, L p 、ap and b p are the different color values ​​of pixels in the color space, and are the different color values ​​of the seed point in the color space, D L , D a and D b They are the color coordinate weights for different colors respectively.

[0023] Optionally, fundus blood vessels are identified in the segmented area to obtain identification results, including:

[0024] Extracting regional features from the segmented area, and selecting the optic disc area according to the regional features;

[0025] The optic disc edge of the optic disc area is obtained by threshold segmentation method, and the optic disc diameter is obtained according to the optic disc edge;

[0026] The blood vessels in the segmented area are classified by blood vessel color to obtain the arteriovenous blood vessel classification results;

[0027] The blood vessels are skeletonized and the window is slid to obtain the arteriovenous diameters, and the arteriovenous ratio is obtained through the arteriovenous diameters.

[0028] Optionally, the training process of the fundus blood vessel segmentation model includes:

[0029] The historical fundus images are input as training sets into the initial model for feature extraction to obtain identification features; the identification features include: color features, texture features, edge features, shape features and intensity features;

[0030] The identification features are fused to obtain fused features, and the fused features are standardized to obtain integrated features;

[0031] The integrated features are optimized through the loss function to obtain the training output;

[0032] The historical fundus images are manually annotated to obtain annotated images, and the annotated images are used as validation sets to verify the training outputs to obtain a fundus vascular segmentation model.

[0033] Optionally, the initial model includes a multi-scale convolution layer, a hole convolution layer and a feature reconstruction layer connected in sequence; the multi-scale convolution layer is composed of multiple convolution layers, and different convolution layers have convolution kernels of different sizes built in; the feature reconstruction layer has a built-in deconvolution algorithm.

[0034] Optionally, the loss function is expressed as: Among them, L is the loss function, N is the total number of samples, y is the true label, p is the predicted probability, ω is the sample weight, λ is the coefficient weight, and i is the i-th pixel.

[0035] An intelligent fundus blood vessel segmentation system based on retinal angiography, comprising:

[0036] An image acquisition module, used for acquiring fundus images by retinal angiography technology;

[0037] An image processing module is used to perform a preprocessing operation on the fundus image to obtain a preprocessed image;

[0038] An image segmentation module is used to perform superpixel segmentation on the preprocessed image to obtain segmented areas;

[0039] An image recognition module is used to identify fundus blood vessels in the segmented area and obtain a recognition result;

[0040] The model building module is used to build a trained fundus vascular segmentation model, and perform feature segmentation on the recognition result through the fundus vascular segmentation model to obtain the segmentation result.

[0041] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the intelligent segmentation method of fundus blood vessels based on retinal angiography provided by the present invention comprises: acquiring a fundus image through retinal angiography technology; performing a preprocessing operation on the fundus image to obtain a preprocessed image; performing superpixel segmentation on the preprocessed image to obtain a segmented area; performing fundus blood vessel recognition on the segmented area to obtain a recognition result; constructing a trained fundus blood vessel segmentation model, and performing feature segmentation on the recognition result through the fundus blood vessel segmentation model to obtain a segmentation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 It is a flow chart of the intelligent segmentation method of fundus vascular of the present invention;

[0044] Figure 2 It is a flow chart of the pretreatment operation of the present invention;

[0045] Figure 3 It is a superpixel segmentation flow chart of the present invention;

[0046] Figure 4 is a flow chart of fundus blood vessel identification of the present invention;

[0047] Figure 5 This is a training flowchart of the fundus blood vessel segmentation model of the present invention. DETAILED DESCRIPTION

[0048] 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.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1 As shown, the present invention provides a method for intelligent segmentation of fundus blood vessels based on retinal angiography, comprising the following steps:

[0051] Step 100: Acquire a fundus image by retinal angiography technology; the fundus image includes: a full-field image and a local enhanced image;

[0052] Specifically, in the process of acquiring fundus images using retinal angiography (FFA) technology, the acquired images are recorded in real time to ensure high-quality fundus retinal images, which contain distinct vascular structures, optic discs, and surrounding anatomical features. The full-field image can cover most of the fundus area, while the local enhanced image focuses on capturing the local area with dense blood vessels. Each image needs to be taken at different time points to capture the dynamic changes and details of the fundus blood vessels.

[0053] More specifically, the size of the local enhanced image is usually between 512*512 pixels and 1024*1024 pixels, depending on the retinal imaging device used and its resolution. The size of the local enhanced image in this embodiment is 1024*1024 pixels. The number of blood vessels contained in the local enhanced image is generally between dozens and hundreds, depending on the specific area analyzed and the complexity of the blood vessels. In some clearer and more vascular areas, approximately 100 to 300 blood vessels may be observed. In smaller local areas, especially when the focus is on major arteries and veins, the number may be between 20 and 50. This embodiment determines an image with more than 20 blood vessels as a local enhanced image.

[0054] It should be noted that by recording fundus images in real time, high-quality retinal images can be ensured, and the distinct vascular structure and anatomical features enable doctors to observe and analyze fundus conditions more clearly. The full-field image can cover most of the fundus area, helping doctors to understand the overall condition of the fundus more comprehensively, while the local enhanced image can focus on the local area with dense blood vessels, which is convenient for identifying potential lesions or abnormalities. In addition, the high resolution (1024*1024 pixels) of the local enhanced image makes subtle vascular changes and dynamic features more significant, enhancing the ability to detect fundus lesions early. By comparing images taken at different time points, doctors can observe the dynamic changes of blood vessels, providing an important basis for the progression of the disease and the evaluation of treatment effects. It can also effectively capture the number and complexity of blood vessels in a specific area, allowing doctors to make more accurate diagnoses based on the number and distribution of blood vessels during analysis, thereby improving the treatment effect of patients and providing strong technical support for the diagnosis and treatment of fundus diseases.

[0055] Step 200: preprocess the fundus image to obtain a preprocessed image. Figure 2 As shown, including:

[0056] Step 201: De-noising the fundus image by using a Gaussian filter to obtain a de-noised image;

[0057] Specifically, filtering techniques (such as Gaussian filtering or median filtering) are used to reduce random noise in fundus images to improve image clarity. This embodiment uses a Gaussian filter to perform weighted averaging of the pixel values ​​of each pixel and its surrounding neighborhood through a convolution operation, so that low-frequency information is retained and high-frequency noise in the image is suppressed, thereby generating a denoised image, while ensuring the characteristics of the vascular structure and reducing the impact of background noise.

[0058] Step 202: performing histogram equalization processing on the denoised image to obtain an enhanced image;

[0059] Specifically, histogram equalization redistributes the grayscale values ​​of pixels in the image to make the brightness distribution in the image more uniform, especially between dark and bright areas. This improves the contrast of the image, making the difference between blood vessels and background more obvious, and improving the clarity of visual information. It also ensures that the details of smaller blood vessels are also preserved, making the vascular features in the retina more prominent.

[0060] Step 203: binarize the enhanced image to obtain a preprocessed image.

[0061] Specifically, a fixed threshold or adaptive thresholding technique is used to divide pixels into foreground (usually blood vessels) and background parts in order to highlight the blood vessel area. The optimal threshold in the processing process is determined by the Otsu algorithm.

[0062] It should be noted that Gaussian filtering effectively removes noise from the image, providing a clean foundation for subsequent image analysis. Histogram equalization enhances the contrast of the image, making details and features more obvious, which can help doctors identify potential lesions more quickly. Through binarization processing, key structural features can be systematically extracted, making subsequent analysis more focused and efficient. This series of preprocessing operations not only improves the clarity and readability of the image, but also provides important technical support for subsequent automated analysis, improving the detection rate and diagnostic accuracy of fundus diseases.

[0063] Step 300: Perform superpixel segmentation on the preprocessed image to obtain segmented regions. The specific steps are as follows: Figure 3 As shown, including:

[0064] Step 301: Determine the number of superpixels K according to the size of the preprocessed image;

[0065] Specifically, the number of superpixels K determines the segmentation accuracy of the image. If the number of superpixels specified is too large, it will lead to insufficient detail features in some areas. The K value of this embodiment is 100.

[0066] Step 302: obtaining a superpixel size according to the number of superpixels, and determining a seed point according to the superpixel size;

[0067] Specifically, the calculation formula for superpixel size is: Where S is the expected size of each superpixel and N is the total number of pixels in the image. Then K seed points are uniformly selected on a segmented area of ​​the preprocessed image, and the positions of the seed points are selected at intervals of S. These seed points will determine the center of the superpixel at the initial moment.

[0068] Step 303: extracting seed vectors of the seed points, and performing superpixel allocation according to the distances between the seed vectors to obtain an allocation result;

[0069] Specifically, the feature vector of each seed point (x, y) is calculated, including its spatial position and color information. The feature vector D is expressed as D = (x, y, L, a, b), where L, a, and b are the color values ​​of the pixel in the CIE LAB color space. Then, the distance from each pixel p in the image to each seed point c is calculated based on the comprehensive metric of spatial distance and color distance. k The distance Dis is calculated as follows:

[0070]

[0071] Among them, (x p ,y p ) is the position of pixel p, is the seed point c k Location, L p 、a p and b p are the different color values ​​of pixels in the color space, and are the different color values ​​of the seed point in the color space, D L , D a and D b are the color coordinate weights of different colors, which are used to balance the influence of different dimensions. L +D a +D b = 1. Finally, according to the calculated distance, each pixel p is assigned to its nearest seed point c k .

[0072] Step 304: Update the superpixel according to the allocation result to obtain the segmented area.

[0073] Specifically, the color values ​​and positions of all pixels in each superpixel are counted, and the new seed point is calculated The calculation formula for the new seed point is:

[0074]

[0075] Among them, N k For c k Repeat step 303 until the center change of all superpixels is less than a predetermined threshold or the preset maximum number of iterations is reached. Finally, the superpixels of the new seed point are checked, and superpixels with too small areas are merged to obtain the segmented area.

[0076] It should be noted that by specifying an appropriate number of superpixels K, the accuracy and effect of segmentation are ensured and the accuracy and detail of image segmentation are balanced. By evenly selecting seed points, it can be ensured that different regions can be effectively segmented and represented, making the segmentation results more representative, thereby improving the stability and consistency of the results. By combining spatial features and color information to calculate the distance between each pixel and the seed point, the risk of misclassification caused by a single feature parameter is reduced, and the classification accuracy is improved. Through iterative updates and superpixel merging, the segmentation results can be continuously optimized, the accuracy of the segmented area is enhanced, and the processing efficiency is further improved, so that the final segmented area is more in line with the needs of medical image analysis.

[0077] Step 400: Identify the fundus blood vessels in the segmented area to obtain the identification results; the identification results include: optic disc diameter, arteriovenous blood vessel classification results and arteriovenous ratio; the specific steps are as follows Figure 4 As shown, including:

[0078] Step 401: extracting regional features from the segmented regions, and selecting the optic disc region according to the regional features;

[0079] Specifically, the boundary features and color distribution features of different regions are analyzed and extracted using region growing or edge detection methods. Based on the extracted features, a classifier based on machine learning is used to identify possible optic disc regions. The identification conditions include that the shape of the region has circular features and the color distribution features conform to the typical features of the optic disc (such as a brighter circle).

[0080] Step 402: obtaining the optic disc edge of the optic disc area by threshold segmentation method, and obtaining the optic disc diameter according to the optic disc edge;

[0081] Specifically, the threshold is dynamically calculated according to the regional features by the Otsu method, and the optimal threshold is automatically determined by the grayscale histogram. The segmented area is divided into foreground (optic disc area) and background according to the optimal threshold. The Canny edge detection algorithm is then used to calculate the gradient of the image to find the edge position. Finally, the images containing the edge position are spliced ​​together to obtain the optic disc edge.

[0082] More specifically, the edge contour of the optic disc is identified through the contour extraction algorithm, and the contour with the largest area is selected as the edge contour of the optic disc. The minimum circumscribed circle of the selected contour is calculated to obtain the center and radius information, and then all points on the contour are traversed to obtain the optic disc diameter by calculating the distance between the two farthest points covered by the contour.

[0083] Step 403: classify the segmented area by blood vessel color to obtain arteriovenous blood vessel classification results;

[0084] Specifically, arteries and veins are distinguished based on the intensity of their color, hue, and diameter. Arteries are usually thinner while veins are thicker; arteries are bright red while veins are dark red, so arteries are usually brighter than veins.

[0085] Step 404: skeletonize the blood vessels and slide the windows to obtain the diameters of the arteries and veins, and obtain the arteriovenous ratio through the diameters of the arteries and veins.

[0086] Specifically, the classified arteries and veins are skeletonized to extract the skeleton of the blood vessels, thereby simplifying the structure and retaining important shape information. The thinning algorithm is then used to normalize the blood vessel width, and the skeleton line obtained after normalization will provide the blood vessel centerline. Then, the fixed window sliding technique is used to slide the blood vessel skeleton, and the arterial diameter and vein diameter in the window are compared to obtain the arteriovenous ratio.

[0087] It should be noted that by combining regional feature extraction technology with threshold segmentation technology, the optic disc area and its edge can be effectively distinguished, ensuring accurate measurement of the optic disc diameter, and providing a reliable basis for subsequent data analysis. The accurate classification of arteriovenous vessels is achieved through vascular color analysis, providing doctors with intuitive indicators for evaluating vascular health status. The calculation of arteriovenous diameter and arteriovenous ratio provides an important quantitative basis for monitoring the health status and pathological changes of fundus blood vessels. This improves the accuracy of image detection and enhances the reliability of automated analysis.

[0088] Step 500: construct a trained fundus vascular segmentation model, and perform feature segmentation on the recognition result through the fundus vascular segmentation model to obtain a segmentation result. The training process of the fundus vascular segmentation model includes: feature extraction, feature integration and loss function optimization.

[0089] Specifically, the training process of the fundus vessel segmentation model is as follows: Figure 5 As shown, including:

[0090] Step 501: Input the historical fundus images as training sets into the initial model to extract features and obtain identification features. Identification features include: color features, texture features, edge features, shape features and intensity features.

[0091] Specifically, the initial model includes a multi-scale convolution layer, a hole convolution layer, and a feature reconstruction layer connected in sequence. The multi-scale convolution layer is composed of multiple convolution layers. The multi-scale characteristics are added to the first few layers of the model, and the image features are processed in parallel through multiple convolution kernels of different sizes, thereby providing multi-level feature extraction from local to global. The hole convolution layer allows the model to have a more thorough understanding of the length and branches of blood vessels without increasing the number of parameters. The feature reconstruction layer is located at the end of the model. It reconstructs the vascular features through deconvolution technology to generate high-definition images; feature reconstruction utilizes the upstream feature information and combines the features of different levels with different weights to reconstruct the segmentation results.

[0092] More specifically, a large number of historical fundus images, including normal and abnormal fundus images, are collected as training sets for the model. Ensure that the training set is diverse, covering different patients, different disease conditions, and different imaging conditions. The historical fundus images are converted using the HSV color space to obtain information on the three channels of hue, saturation, and brightness. Color features are extracted using color histogram analysis and statistics such as mean and variance in the color space. Threshold segmentation is used to extract vascular areas, determine pixels within a specific range of hue values, and mark them as potential vascular areas.

[0093] Gabor filters and local binary patterns (LBP) are used to obtain texture features such as contrast, uniformity, and entropy of fundus images. Gabor filters can effectively capture the subtle texture details of blood vessels through responses in different directions and scales. Using these features enables the model to better recognize the thickness, distortion, and branching characteristics of blood vessels.

[0094] The Canny edge detection algorithm is used to extract edge features in fundus images and generate contour lines to enhance the boundary definition of the region. The algorithm first performs Gaussian filtering, then calculates the gradient and performs non-maximum suppression, and finally uses a double threshold method to determine the edge. Edge features help find the contours of the retina and blood vessels, which is crucial for the accurate positioning and classification of structures, especially providing clear boundary information for segmentation and subsequent analysis.

[0095] The Hough transform technique is used to detect the centerline of the blood vessel and extract the contour features (such as perimeter, area, shape factor, etc.) and shape descriptors. The Hough transform is used to detect straight lines and circles in the image and can extract the geometric shape information of the blood vessels, further helping to establish the model of the blood vessels and their distribution characteristics. Shape features are used to analyze the geometric shape of structures in fundus images, which can help identify the features of the optic disc, blood vessels, and abnormal areas, and provide information that is helpful for disease diagnosis.

[0096] Calculate the grayscale mean, standard deviation, and maximum and minimum grayscale values ​​of the image to obtain the image brightness distribution, and use this to analyze the intensity changes in different areas. The intensity feature reflects the overall illumination and contrast of the image, which can help supplement the color and texture features and further enhance the classification ability of the features.

[0097] Step 502: Fusing the identified features to obtain fused features, and standardizing the fused features to obtain integrated features.

[0098] Furthermore, the features extracted separately (color, texture, edge, shape and intensity features) are integrated to obtain fused features and expressed in the form of feature vector F, which is expressed as:

[0099] F=α(Y⊙W⊕B)+(β·Y+γ·X)⊕Q;

[0100] Among them, α is the weight coefficient calculated by genetic algorithm, β and γ are the adjustment coefficients obtained through training, which are used to dynamically adjust the influence of different features, β+γ=1, Y is color feature, W is texture feature, B is edge feature, X is shape feature, and Q is intensity feature. Then the feature vector is processed by Z-score normalization method to ensure that all features are in the same range.

[0101] Step 503: Optimize the loss of the integrated features through the loss function to obtain the training output.

[0102] Specifically, the loss function is expressed as:

[0103]

[0104] Among them, L is the loss function, N is the total number of samples, y is the true label, p is the predicted probability, ω is the sample weight, λ is the coefficient weight, and i is the i-th pixel.

[0105] Step 504: manually annotate the historical fundus images to obtain annotated images, and use the annotated images as a verification set to verify the training output to obtain a fundus blood vessel segmentation model.

[0106] The present invention also provides a fundus blood vessel intelligent segmentation system based on retinal angiography, comprising:

[0107] An image acquisition module, used for acquiring fundus images by retinal angiography technology;

[0108] An image processing module is used to perform a preprocessing operation on the fundus image to obtain a preprocessed image;

[0109] An image segmentation module is used to perform superpixel segmentation on the preprocessed image to obtain segmented areas;

[0110] An image recognition module is used to identify fundus blood vessels in the segmented area and obtain a recognition result;

[0111] The model building module is used to build a trained fundus vascular segmentation model, and perform feature segmentation on the recognition result through the fundus vascular segmentation model to obtain the segmentation result.

[0112] The beneficial effects of the present invention are as follows:

[0113] 1) A multi-stage image processing process is adopted, including image acquisition, preprocessing, superpixel segmentation, and blood vessel identification. Each stage is optimized for a specific task, which improves the accuracy of the overall processing, reduces the need for manual intervention, and improves the efficiency of fundus image analysis;

[0114] 2) Using Gaussian filter to reduce noise and histogram equalization processing ensures that the image has higher contrast and clarity during segmentation and recognition, greatly reducing the interference of noise on subsequent analysis;

[0115] 3) Through the superpixel algorithm, more accurate regional segmentation is performed based on image features and seed points, which can retain more detailed features and ensure the accuracy of subsequent further analysis of blood vessels and optic disc areas;

[0116] 4) Dynamically calculate the threshold through the Otsu method and combine it with Canny edge detection to ensure that the optic disc edge and other key structures can be accurately extracted, thereby improving the accuracy of optic disc diameter and blood vessel classification;

[0117] 5) Feature extraction includes color, texture, edge, shape and intensity features, which can fully reflect image information from multiple dimensions, provide sufficient feature information for training models, and enhance the learning and generalization capabilities of the model;

[0118] 6) The initial model uses multi-scale convolutional layers and dilated convolutional layers to process image features in different ways, which can adapt to the complexity of various retinal images and ensure that the extracted vascular features are rich and accurate.

[0119] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0120] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for intelligent segmentation of fundus vessels based on retinal angiography, characterized in that: The steps include: Obtaining fundus images by retinal angiography technology; the fundus images include: full-field images and local enhanced images; Performing a preprocessing operation on the fundus image to obtain a preprocessed image; Performing superpixel segmentation on the preprocessed image to obtain segmented areas; Perform fundus blood vessel recognition on the segmented area to obtain recognition results; the recognition results include: optic disc diameter, arteriovenous blood vessel classification results and arteriovenous ratio; A trained fundus vascular segmentation model is constructed, and feature segmentation is performed on the recognition result through the fundus vascular segmentation model to obtain a segmentation result; the training process of the fundus vascular segmentation model includes: feature extraction, feature integration and loss function optimization.

2. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 1, characterized in that: The full-field image is an image covering the entire fundus range; the size of the local enhanced image is 1024*1024 pixels, and the number of blood vessels in the local enhanced image is greater than 20.

3. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 1, characterized in that: Performing a preprocessing operation on the fundus image to obtain a preprocessed image includes: Performing denoising on the fundus image by using a Gaussian filter to obtain a denoised image; Performing histogram equalization processing on the denoised image to obtain an enhanced image; The enhanced image is binarized to obtain the preprocessed image.

4. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 1, characterized in that: Performing superpixel segmentation on the preprocessed image to obtain segmented regions includes: Determining the number of superpixels according to the size of the preprocessed image; Obtaining a superpixel size according to the number of superpixels, and determining a seed point according to the superpixel size; Extracting seed vectors of the seed points, and performing superpixel allocation according to distances between the seed vectors to obtain an allocation result; The superpixel is updated according to the allocation result to obtain the segmented area.

5. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 4, characterized in that: The calculation formula of the distance between the seed vectors is: Where Dis is the vector distance, (x p ,y p ) is the position of pixel p, is the seed point c k Location, L p 、a p and b p are the different color values ​​of pixels in the color space, and are the different color values ​​of the seed point in the color space, D L , D a and D b They are the color coordinate weights for different colors respectively.

6. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 1, characterized in that: Perform fundus blood vessel recognition on the segmented area to obtain a recognition result, including: Extracting regional features from the segmented regions, and selecting the optic disc region according to the regional features; Obtaining the optic disc edge of the optic disc area by a threshold segmentation method, and obtaining the optic disc diameter according to the optic disc edge; Classify the blood vessels in the segmented area according to the blood vessel color to obtain the arteriovenous blood vessel classification result; The blood vessels are subjected to skeletonization and window sliding to obtain arteriovenous diameters, and the arteriovenous ratio is obtained through the arteriovenous diameters.

7. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 1, characterized in that: The training process of the fundus blood vessel segmentation model includes: Inputting historical fundus images as training sets into the initial model to extract the features and obtain identification features; the identification features include: color features, texture features, edge features, shape features and intensity features; Performing feature fusion on the identification features to obtain fused features, and performing standardization processing on the fused features to obtain integrated features; Performing loss optimization on the integrated features through the loss function to obtain a training output; The historical fundus images are manually annotated to obtain an annotated image, and the annotated image is used as a verification set to verify the training output to obtain the fundus blood vessel segmentation model.

8. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 7, characterized in that: The initial model includes a multi-scale convolution layer, a hole convolution layer and a feature reconstruction layer connected in sequence; the multi-scale convolution layer is composed of multiple convolution layers, and different convolution layers have convolution kernels of different sizes built in; the feature reconstruction layer has a deconvolution algorithm built in.

9. The method for intelligent segmentation of fundus vessels based on retinal angiography according to claim 7, characterized in that: The expression of the loss function is: Among them, L is the loss function, N is the total number of samples, y is the true label, p is the predicted probability, ω is the sample weight, λ is the coefficient weight, and i is the i-th pixel.

10. An intelligent fundus blood vessel segmentation system based on retinal angiography, characterized in that: include: An image acquisition module, used for acquiring fundus images by retinal angiography technology; An image processing module, used for performing a preprocessing operation on the fundus image to obtain a preprocessed image; An image segmentation module, used to perform superpixel segmentation on the preprocessed image to obtain segmented areas; An image recognition module, used to identify fundus blood vessels in the segmented area to obtain a recognition result; The model building module is used to build a trained fundus vascular segmentation model, and perform feature segmentation on the recognition result through the fundus vascular segmentation model to obtain a segmentation result.

Images