A method for ankle talus cartilage segmentation and thickness measurement

Abstract

The application provides a kind of ankle talus cartilage segmentation and thickness measurement method, it is related to image processing technical field, the ankle talus nuclear magnetic image of the present application is collected, constructs ankle talus cartilage segmentation standard data set to be used to network training. Method first to ankle talus nuclear magnetic image is standardized, and the data set of being labeled is used to train segmentation model. Then, the trained model is applied to new nuclear magnetic image data, and the segmentation result of talus cartilage is obtained by model prediction. Finally, based on the segmentation result, the average thickness of talus cartilage is calculated using mathematical algorithms and image processing techniques. The method provided by the present application can provide accurate diagnostic basis for clinicians, help orthopedic surgeons predict the amount of cartilage that needs to be repaired, avoid over-treatment, significantly improve the accuracy and efficiency of ankle talus cartilage segmentation, and reduce the time and error of manual operation. The method has wide application prospects in talus cartilage injury assessment and treatment effect monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a method for ankle talus cartilage segmentation and thickness measurement. BACKGROUND

[0002] The health status of ankle talus cartilage is crucial in clinical diagnosis and treatment. By measuring the thickness of talus cartilage, not only can orthopedic surgeons predict the amount of cartilage that needs to be repaired, but also can avoid over-treatment and reduce the harm to patients.

[0003] In the prior art, ultrasound and magnetic resonance imaging (MRI) are commonly used methods for cartilage measurement. Both are safe and harmless without radiation, but studies have shown that MRI is superior to ultrasound in the accuracy of true cartilage measurement, and therefore has become the best choice for indirect observation of cartilage.

[0004] Current cartilage thickness measurement methods mainly rely on manual operation, which has the problems of low efficiency and subjective error. In addition, studies have found that cartilage thickness has no direct correlation with the length and thickness of the skeleton, so it cannot be simply predicted by skeletal CT or X-ray film.

[0005] Deep learning technology has shown great potential in the field of medical image processing in recent years, especially in image segmentation and feature extraction. Semantic segmentation methods based on deep learning can automatically learn and identify cartilage tissue in images based on a large amount of labeled data, significantly improving the accuracy and efficiency of segmentation. However, there is no systematic research report on the measurement of ankle cartilage thickness in China. SUMMARY

[0006] In view of the deficiencies of the prior art, the purpose of the present application is to provide a method for ankle talus cartilage segmentation and thickness measurement.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] A method for ankle talus cartilage segmentation and thickness measurement, comprising the following steps:

[0009] S1, collecting human ankle talus nuclear magnetic images and performing data preprocessing;

[0010] First, the ankle talus nuclear magnetic images are exported from the hospital. The ankle talus nuclear magnetic images used are original resolution 512 512 ankle MR plain scan images, picture format is jpg;

[0011] Then, the talus cartilage layer in the nuclear magnetic image is screened, and the nuclear magnetic image in which the talus cartilage is clearly visible is retained; the ankle talus nuclear magnetic image used is limited by hospital equipment and shooting environment, and part of the nuclear magnetic image cannot clearly display the talus cartilage layer of the ankle joint, so it is necessary to manually screen and retain the nuclear magnetic image in which the talus cartilage is clearly visible;

[0012] Finally, the nuclear magnetic image is uniformly cropped to retain the talus cartilage part and make the talus cartilage in the center of the picture, and the final image resolution used for making the ankle talus cartilage segmentation data set is 160 160, and the picture format is png;

[0013] S2, label the ankle talus cartilage segmentation label, and perform data enhancement and data set division;

[0014] First, the ankle talus cartilage segmentation label is labeled according to the contour of the talus cartilage, a single-channel binary image containing the ankle talus cartilage region is generated, and single-channel means that the image has only one gray channel. The value of each pixel is usually an integer between 0 and 255, representing the brightness or gray value of the pixel. Therefore, each pixel is represented by a value indicating its position on the gray scale. Binary means that each pixel has only two possible values, corresponding to the presence and absence of a certain property in the image. Such an image can simplify analysis and processing, and this image is used as the label of the talus cartilage segmentation data set; the pixel value of the foreground talus cartilage region is 1, and the pixel value of the background non-talus cartilage region is 0;

[0015] Then, multiple data enhancement methods are used to expand the sample size of the ankle talus cartilage segmentation data set;

[0016] The data enhancement methods include: mirror flipping along the vertical axis of the image, which helps the model learn the performance of the object in different directions; random-15°~+15° angle rotation, which enhances the model's ability to adapt to the rotation of the object, making it better handle the performance of the object at different angles; random 0.8~1.2 times scaling, randomly selecting a scaling ratio between 0.8 and 1.2, so that the model can handle objects of different scales; scale transformation, shear transformation, which stretches part of the image along the horizontal or vertical direction to form a parallelogram, which can simulate the deformation of the object when the viewing angle changes; Gaussian noise, adding random noise obeying Gaussian distribution to the pixel value of the image. By introducing noise, the model is more likely to learn the characteristics of real-world data, improving the generalization ability; Gaussian blur, Gaussian blur processing of the image, the blur degree is controlled by a standard deviation parameter, which can reduce the details and noise in the image, helping the model focus on the overall shape and structure of the object;

[0017] Finally, the training set and the test set of the ankle talus cartilage segmentation dataset are randomly divided in a ratio of 7:3 of the number of images;

[0018] S3, constructing an ankle talus cartilage segmentation network model, using the labeled data to train the model;

[0019] First, the feature extraction part in the network model is constructed; the feature extraction part is composed of a multi-scale deep separable convolution block (MDSC block); the feature extraction part is composed of five stages, generating from lower-level abstract features to higher-level semantic features;

[0020] The core of the feature extraction part is the multi-scale deep separable convolution block (MDSC block). First, the input channel number is adjusted through the max-pooling layer. The max-pooling layer operation can basically be understood as selecting the maximum value as the output in each local region (for example, a window of 2 2) of the input, which is used to reduce the spatial size of the feature map while retaining important features; then, the features are extracted by the multi-scale deep separable convolution, which is a special type of convolution operation in convolutional neural networks (CNN) that can effectively reduce the number of model parameters and computational complexity while improving the performance of the model to some extent. The convolution extracts features in parallel by five different size deep convolutions, fuses the extracted features, and then aligns the channel number using point-wise convolution. Then, the batch normalization layer standardizes the output of the multi-scale deep separable convolution, which is a commonly used technique in deep learning that can accelerate the training process of neural networks and improve the convergence speed and stability of the model. Finally, ReLU is used as the activation function;

[0021] The ReLU activation function is used for nonlinear mapping of the convolution output features. One implementation is shown in equation (1):

[0022]

[0023] Then, the talus cartilage segmentation part is constructed; the talus cartilage segmentation part is composed of a KAN enhanced multi-scale upsampling block (KEMU block); the talus cartilage segmentation part is used to generate pixel-level talus cartilage segmentation results;

[0024] The talus cartilage segmentation part fuses the upsampled feature map with the feature map of the corresponding level through skip connections, mapping the extracted multi-level features back to the original image size layer by layer. The final layer of the talus cartilage segmentation network generates pixel-level segmentation results. Part of the composition of the talus cartilage segmentation part is similar to that of the feature extraction network, but the difference lies in the introduction of KAN linear layers and channel spatial attention from the Kolmogorov-Arnold Networks (KAN). The KAN linear layers, as enhanced linear layers, provide stronger non-linear representation capabilities; channel spatial attention enhances the feature map by progressively applying channel attention and spatial attention.

[0025] S4. Construct the loss function, determine the optimizer and hyperparameters, input the training set, and train the network parameters;

[0026] First, the model is initialized, the dataset is loaded, and data preprocessing is performed. The dataset is in Pascal VOC (Visual Object Classes) format; data preprocessing includes randomly resizing images to a specified range (200). 200-300 300), horizontally flip, randomly crop the image, and normalize the pixel values ​​of the image. Adjust the pixel values ​​of the image according to certain rules so that the different parts of the image have similar data distribution or statistical characteristics; the feature extraction part extracts multi-level features of the original ankle joint talus MRI image, and the talus cartilage segmentation part segments the talus cartilage to obtain the talus cartilage segmentation prediction map.

[0027] Then, the segmentation loss of the talus cartilage is calculated, which is the loss between the predicted segmentation map of the talus cartilage and the corresponding segmentation label of the talus cartilage.

[0028] The loss function consists of Dice loss and binary cross-entropy loss, and one implementation of it is shown in equation (2):

[0029]

[0030] In equation (2), Real labels were segmented for each talus cartilage. This is the corresponding segmentation prediction map of the talus cartilage. This represents the total number of images in the training set. Use the smallest positive integers to ensure that the denominator is not zero;

[0031] Finally, the gradient is calculated using the backpropagation algorithm, and the optimizer is used to update the model's parameters;

[0032] The optimizer is the AdamW optimizer, and its parameter update strategy is defined by equation (3):

[0033]

[0034] In equation (3), In the first The parameters at the next iteration and These are the first and second momentum of the gradient, respectively. This is the learning rate, set to 0.001. It is a very small constant, set to 1e-8. This is the weight decay coefficient, set to 0.01;

[0035] S5. Load the trained weight parameters into the constructed segmentation model, input the test set data into the model, perform image post-processing on the output image, and obtain the segmentation result of the talus cartilage.

[0036] First, load the trained model weight parameters and input the test set data into the model; then, perform image post-processing on the output image to convert the output binary image into an RGB image; finally, obtain the segmentation result image of the talus cartilage.

[0037] S6. Use the segmentation results to calculate the average thickness of the talus cartilage.

[0038] First, the segmentation result image of the talus cartilage is loaded, and the image is processed into a binary image; the segmentation result image is a grayscale image, and the grayscale image is converted into a binary image. This process will set pixels with grayscale values ​​greater than the threshold to white (255) and pixels with grayscale values ​​less than or equal to the threshold to black (0) according to the threshold.

[0039] Then, the conversion coefficient between pixels and actual distance is calculated; the number of pixels per centimeter is calculated using a centimeter ruler in the original image, and then converted to actual distance using the conversion coefficient. The original image size is 512. 512 pixels. The measured image is 160. 160 pixels, therefore it is necessary to calculate the ratio between the actual distance and the pixel based on the scaling ratio, and convert the pixel value into the actual physical distance value;

[0040] Next, contours are detected in the binary image; contours represent continuous white areas in the image, and each contour may correspond to a region of the talus cartilage.

[0041] Finally, the average thickness of the two talar cartilage segments within the contour region is calculated and converted into an actual thickness value. For each detected contour, a mask image of the same size as the original image is created. First, the contour is filled with white, and the rest with black. Then, the mask image is traversed row by row, calculating the number of white pixels in each row; this number represents the thickness of the talar cartilage. The average thickness of the region is obtained by averaging the thicknesses of all rows.

[0042] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0043] (1) This invention provides a unique and specially designed dataset for segmentation of the talus cartilage of the ankle joint, which solves the problem of insufficient existing data and provides necessary resources and foundation for future related research and applications.

[0044] (2) The present invention provides a segmentation network model for ankle joint talus cartilage, which significantly improves the accuracy and efficiency of ankle joint cartilage segmentation tasks.

[0045] (3) This invention provides a method for measuring the thickness of the talus cartilage in the ankle joint, offering physicians a faster and more accurate tool for assessing cartilage health. This technological innovation helps improve the diagnostic process and enhances patient treatment outcomes and health management. Attached Figure Description

[0046] Figure 1 This is a flowchart of the present invention;

[0047] Figure 2 This is a schematic diagram of the segmentation network model framework of the present invention;

[0048] Figure 3 This refers to the multi-scale depthwise separable convolutional block (MDSC block) in the segmentation network model of this invention.

[0049] Figure 4 This is the KAN-enhanced multi-scale upsampling block (KEMU block) in the segmentation network model of this invention.

[0050] Figure 5 The images shown are the test results of talus cartilage segmentation using the ankle joint talus cartilage segmentation network model in this embodiment of the invention, and the results of measuring the thickness of the ankle joint talus cartilage using the segmentation result images. In this embodiment, (a) is the input ankle joint talus cartilage segmentation dataset test image, (b) is the corresponding ground truth label for talus cartilage segmentation, (c) is the corresponding talus cartilage segmentation test mask, (d) is the average thickness measured by the corresponding ground truth label for talus cartilage segmentation, and (e) is the average thickness measured by the corresponding talus cartilage segmentation test mask. Detailed Implementation

[0051] Reference Figures 1 to 5 The embodiments of the present invention will be described in further detail below. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the invention.

[0052] A method for segmenting and measuring the thickness of the talus cartilage in the ankle joint, such as Figure 1 As shown, it includes the following steps:

[0053] Step 1: Acquire MRI images of the talus at the ankle joint and perform data preprocessing;

[0054] Specifically, this involves exporting MRI images of the talus bone of the ankle joint from a hospital. In this embodiment, the MRI images of the talus bone of the ankle joint used have an original resolution of 512. MR plain scan of the ankle joint of 512, image format is jpg;

[0055] The talus cartilage layer in the MRI images was screened to retain clear images. In this embodiment, the ankle talus MRI images used were limited by hospital equipment and the imaging environment; some images did not clearly show the talus cartilage layer of the ankle joint, requiring manual screening to retain MRI images where the talus cartilage was clearly visible.

[0056] Unified cropping of MRI images. This embodiment describes a method for segmenting and measuring the thickness of the talus cartilage in the ankle joint. The original MRI images of the talus in the ankle joint contain interference besides the talus cartilage; therefore, unified cropping is necessary to preserve the talus cartilage and center it within the image. The final image resolution used to create the ankle joint talus cartilage segmentation dataset is 160 dpi. 160, image format is PNG;

[0057] Step 2: Label the segmentation of the talus cartilage in the ankle joint and use data augmentation to expand the number of MRI images of the ankle joint to construct an ankle joint talus cartilage segmentation dataset; the ankle joint talus cartilage segmentation dataset consists of ankle joint MRI images and corresponding talus cartilage segmentation labels;

[0058] Specifically as follows:

[0059] Based on the contour of the talus cartilage, the segmentation labels of the talus cartilage of the ankle joint are marked, and a single-channel binary image containing the talus cartilage region of the ankle joint is generated. This image is used as the label of the talus cartilage segmentation dataset; where the pixel value of the foreground talus cartilage region is 1, and the pixel value of the background non-talus cartilage region is 0.

[0060] Multiple data augmentation methods were employed to expand the sample size of the ankle joint talus cartilage segmentation dataset. These data augmentation methods included: mirror flipping along the vertical axis of the image, random rotation at an angle of -15° to +15°, random scaling at a scale of 0.8 to 1.2 times, shear transformation, Gaussian noise, and Gaussian blur.

[0061] The training and test sets of the ankle joint talus cartilage segmentation dataset were randomly divided in a 7:3 ratio of image quantity.

[0062] Step 3: Construct a segmentation network model of the talus cartilage in the ankle joint, and train the model using labeled data. The network model framework is as follows: Figure 2 As shown, specifically:

[0063] The feature extraction part of the network model is constructed; the feature extraction part consists of multi-scale depthably separable convolutional blocks (MDSC blocks); the feature extraction part consists of five stages, generating features from lower-level abstract features to higher-level semantic features;

[0064] In this embodiment, the core component of the feature extraction part is a multi-scale depthably separable convolutional block (MDSC block), with the structure as follows: Figure 3 As shown, first, the number of input channels is adjusted through a max pooling layer; then, features are extracted through a multi-scale depth-separable convolution, which extracts features in parallel from five depth convolutions of different sizes. The extracted features are then fused, and the number of channels is aligned using a pointwise convolution; next, a batch normalization layer normalizes the output of the multi-scale depth-separable convolution; finally, ReLU is used as the activation function.

[0065] The talar cartilage segment is constructed; the talar cartilage segment consists of KAN-enhanced multi-scale upsampling blocks (KEMU blocks), with the structure as follows: Figure 4 As shown. The talus cartilage segmentation portion is used to generate pixel-level talus cartilage segmentation results;

[0066] In this embodiment, the talus cartilage segmentation part fuses the upsampled feature map with the feature map of the corresponding level through skip connections, mapping the extracted multi-level features back to the original image size layer by layer. The last layer of the talus cartilage segmentation network generates pixel-level segmentation results. Part of the composition of the talus cartilage segmentation part is similar to that of the feature extraction network, but the difference lies in the introduction of the KAN linear layer and channel spatial attention from the Kolmogorov-Arnold Network (KAN). The KAN linear layer, as an enhanced linear layer, provides stronger non-linear representation capabilities; channel spatial attention enhances the feature map by progressively applying channel attention and spatial attention, enabling the network to focus more on important features.

[0067] Step 4: Construct the loss function, determine the optimizer and hyperparameters, input the training set, and train the network parameters;

[0068] Specifically as follows:

[0069] Initialize the model, load the dataset, and perform data preprocessing;

[0070] The dataset uses the Pascal VOC (Visual Object Classes) format, which is highly adaptable to various network models; data preprocessing includes randomly resizing images to a specified range (200). 200-300 300), horizontally flipped, randomly cropped, and normalized image pixel values;

[0071] Multi-level features of the original ankle joint talus MRI image were extracted using a feature extraction network, and the talus cartilage was segmented using a talus cartilage segmentation network to obtain a talus cartilage segmentation prediction map.

[0072] Calculate the segmentation loss of the talus cartilage, that is, the loss between the predicted segmentation map of the talus cartilage and the corresponding segmentation label of the talus cartilage;

[0073] The loss function consists of Dice loss and binary cross-entropy loss;

[0074] The gradient is calculated using the backpropagation algorithm, and the model parameters are updated using the optimizer.

[0075] Step 5: Load the trained weight parameters into the constructed segmentation model, input the test set data into the model, perform image post-processing on the output image, and obtain the segmentation result of the talus cartilage;

[0076] Specifically, the process involves: loading the pre-trained model weights and parameters, inputting the test set data into the model; performing image post-processing on the output image, converting the binary image to an RGB image; and obtaining the segmentation result image of the talus cartilage, such as... Figure 5 As shown, (a) is the input test image of the ankle joint talus cartilage segmentation dataset, (b) is the corresponding ground truth label of talus cartilage segmentation, and (c) is the corresponding test mask of talus cartilage segmentation.

[0077] Step 6: Calculate the average thickness of the talus cartilage using the segmentation results;

[0078] Specifically as follows:

[0079] Load the segmentation result image of the talus cartilage and convert the segmentation result image into a binary image. This process will set pixels with gray values ​​greater than the threshold to white (255) and pixels with gray values ​​less than or equal to the threshold to black (0) according to the threshold.

[0080] Calculate the conversion factor between pixels and actual distance. Use a centimeter ruler from the original image to calculate the number of pixels per centimeter, and then use the conversion factor to convert it to the actual distance. The original image size is 512. 512 pixels. The measured image is 160. 160 pixels, therefore it is necessary to calculate the ratio between the actual distance and the pixel based on the scaling ratio, and convert the pixel value into the actual physical distance value;

[0081] Detect contours in a binary image, where contours represent continuous white areas in the image, and each contour may correspond to a region of the talus cartilage;

[0082] The average thickness of the two talar cartilage segments within the contour region is calculated and converted into an actual thickness value. For each detected contour, a mask image of the same size as the original image is created. First, the contour is filled with white, and the rest is black. Then, the mask image is traversed row by row, calculating the number of white pixels in each row; this number represents the thickness of the talar cartilage. The average thickness of the region is obtained by averaging the thicknesses of all rows.

[0083] like Figure 5 As shown, (d) is the average thickness measured by the corresponding talus cartilage segmentation real label, and (e) is the average thickness measured by the corresponding talus cartilage segmentation test mask.

[0084] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for segmenting and measuring the thickness of the talus cartilage in the ankle joint, characterized in that: Includes the following steps: S1. Acquire MRI images of the talus region of the ankle joint, perform image filtering based on the clarity of the talus cartilage and localization and cropping based on the anatomical structure to obtain a preprocessed image with uniform resolution centered on the talus cartilage. S2. Label the talus cartilage region in the preprocessed image and generate binary segmentation labels; expand the dataset by combining multiple enhancement methods, including mirror flipping along the vertical axis, random rotation from -15° to +15°, random scaling from 0.8 to 1.2 times, shear transformation, Gaussian noise and Gaussian blur, and divide the dataset into training set and test set; S3. Construct a segmentation network model of the talus cartilage of the ankle joint. This model includes: - The feature extraction part consists of multi-level, multi-scale, depth-separable convolutional blocks, which are used to extract and fuse features through parallel multi-scale depth convolution;   - The segmentation part consists of KAN-enhanced multi-scale upsampling blocks, which introduce KAN linear layers and channel space attention mechanisms, and perform feature fusion with the corresponding layers of the feature extraction part through skip connections to achieve pixel-level segmentation; S4. Use the weighted sum of the Dice loss function and the binary cross-entropy loss function as the loss function for model training, and use the AdamW optimizer to train the network parameters. S5. Load the weight parameters onto the trained model, perform segmentation prediction on the test set images, and perform binarization and morphological post-processing on the output prediction map to obtain the final talus cartilage segmentation result map. S6. Based on the segmentation result image, the average thickness of the talus cartilage is obtained by using the conversion coefficient between pixels and actual distances and the method of line-by-line contour analysis and average calculation.

2. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 1, characterized in that: Step S1 includes the following: First, MRI images of the talus of the ankle joint were exported from the hospital. The MRI images of the talus of the ankle joint used were original resolution 512. MR plain scan of the ankle joint of 512, image format is jpg; Then, the talus cartilage layer in the MRI images is screened to retain clear MRI images. Due to limitations of hospital equipment and imaging environment, some MRI images of the ankle joint talus cannot clearly show the talus cartilage layer of the ankle joint. Therefore, manual screening is required to retain MRI images in which the talus cartilage is clearly visible. Finally, the MRI images were uniformly cropped to preserve the talus cartilage and center it within the image. The final images used to create the ankle joint talus cartilage segmentation dataset had a resolution of 160. 160, image format is PNG.

3. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 2, characterized in that: Step S2 includes the following: First, the ankle joint talus cartilage segmentation labels are marked according to the contour of the talus cartilage, and a single-channel binary image containing the ankle joint talus cartilage region is generated. This image is used as the label for the talus cartilage segmentation dataset. The pixel value of the foreground talus cartilage region is 1, and the pixel value of the background non-talus cartilage region is 0. Then, various data augmentation methods were used to expand the number of samples in the ankle joint talus cartilage segmentation dataset; the data augmentation methods included: mirror flipping along the vertical axis of the image, random rotation at an angle of -15° to +15°, random scaling at a scale of 0.8 to 1.2 times, shear transformation, Gaussian noise, and Gaussian blur; Finally, the training and test sets of the ankle joint talus cartilage segmentation dataset were randomly divided in a 7:3 ratio of image quantity.

4. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 3, characterized in that: Step S3 includes the following: First, the feature extraction part of the network model is constructed; the feature extraction part consists of multi-scale depthwise separable convolutional blocks; the feature extraction part consists of five stages, generating features from lower-level abstract features to higher-level semantic features; The core of the feature extraction part is a multi-scale depth-separable convolutional block; the number of input channels is adjusted by a max pooling layer; features are extracted by multi-scale depth-separable convolution, which is composed of five depth convolutions of different sizes in parallel. The extracted features are then fused and the number of channels is aligned by pointwise convolution; a batch normalization layer standardizes the output of the multi-scale depth-separable convolution; and ReLU is used as the activation function. Then, a talus cartilage segmentation component is constructed; the talus cartilage segmentation component consists of KAN-enhanced multi-scale upsampling blocks; the talus cartilage segmentation component is used to generate pixel-level talus cartilage segmentation results; The talus cartilage segmentation part uses skip connections to fuse the upsampled feature map with the feature map of the corresponding level, and maps the extracted multi-level features back to the original image size layer by layer. The last layer of the talus cartilage segmentation network generates pixel-level segmentation results. The composition of the talus cartilage segmentation part is similar to that of the feature extraction network, but the difference is that the KAN linear layer and channel space attention from the Kolmogrove-Arnold network are introduced; the KAN linear layer, as an enhanced linear layer, provides stronger nonlinear representation capabilities. Channel spatial attention enhances feature maps by progressively applying channel attention and spatial attention.

5. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 4, characterized in that: Step S4 includes the following: First, the model is initialized, the dataset is loaded, and data preprocessing is performed. The dataset is in Pascal VOC format. Data preprocessing includes randomly resizing images to 200 pixels. 200 to 300 The image is 300 pixels wide, horizontally flipped, randomly cropped, and normalized. The feature extraction part extracts multi-level features from the original ankle joint talus MRI image, and the talus cartilage segmentation part segments the talus cartilage to obtain a talus cartilage segmentation prediction map. Then, the talus cartilage segmentation loss is calculated, which is the loss between the talus cartilage segmentation prediction map and the corresponding talus cartilage segmentation label; the loss is the sum of the Dice loss and the binary cross-entropy loss. Finally, the gradient is calculated using the backpropagation algorithm, and the model parameters are updated using an optimizer; the optimizer is the AdamW optimizer.

6. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 5, characterized in that: Step S5 includes the following: First, load the trained model weight parameters and input the test set data into the model; then, perform image post-processing on the output image to convert the output binary image into an RGB image; finally, obtain the segmentation result image of the talus cartilage.

7. The method for segmenting and measuring the thickness of the talus cartilage in the ankle joint according to claim 6, characterized in that: Step S6 includes the following: First, the segmentation result image of the talus cartilage is loaded, and the image is processed into a binary image; the segmentation result image is a grayscale image, and the grayscale image is converted into a binary image; the process will set pixels with grayscale values ​​greater than the threshold to white and pixels with grayscale values ​​less than or equal to the threshold to black according to the threshold. Then, the conversion coefficient between pixels and actual distance is calculated; the number of pixels per centimeter is calculated using a centimeter ruler in the original image, and then converted to actual distance using the conversion coefficient. The original image size is 512. 512 pixels; the measured image is 160. 160 pixels, therefore it is necessary to calculate the ratio between the actual distance and the pixel based on the scaling ratio, and convert the pixel value into the actual physical distance value; Next, contours are detected in the binary image; contours represent continuous white areas in the image, and each contour may correspond to a region of the talus cartilage. Finally, the average thickness of the two talar cartilages within the contour region is calculated and converted into actual thickness values; for each detected contour, a mask image of the same size as the original image is created. First, fill the outline with white and the rest with black; then, traverse the mask image line by line and calculate the number of white pixels in each line, which represents the thickness of the talus cartilage; by averaging the thickness of all lines, the average thickness of this region is obtained.

Images