Multi-task joint segmentation method for kidney stone and vascular calcification in coronary CTA image

Abstract

The invention discloses a multi-task joint segmentation method for kidney stone and vascular calcification in a coronary CTA image, and relates to the technical field of image segmentation processing, and the segmentation method comprises the following steps: obtaining coronary CTA image data, constructing a multi-task joint segmentation network, inputting the standardized image data into a shared feature encoder, and obtaining a shared feature encoder; and performing inter-task feature interaction enhancement on the multi-scale feature representation through a task interaction module to obtain enhanced feature representation, respectively inputting the enhanced feature representation into a kidney stone segmentation decoder and a vascular calcification segmentation decoder, and generating a final joint segmentation mask image according to a kidney stone segmentation result and a vascular calcification segmentation result. The method has the advantages that the segmentation of kidney stone and vascular calcification is completed in a single network at the same time through a multi-task joint learning framework, and compared with a scheme of independently training two models, the model complexity and computing resource consumption are reduced, and the feasibility of practical application is improved.

Description

Technical Field

[0001] This invention relates to the field of image segmentation processing technology, specifically a multi-task joint segmentation method for kidney stones and vascular calcification in coronary CTA images. Background Technology

[0002] Coronary CT angiography (CTA) is an important imaging examination tool used in clinical practice for the diagnosis of coronary heart disease. During coronary CTA examination, since the scanning range usually covers the area from the chest cavity to the upper abdomen, the images will show not only the target coronary artery structure, but also other tissues and organs, such as the kidney area. Clinical studies have shown that kidney stones and vascular calcification are two common high-density lesions, both of which appear as high-density calcification foci in CTA images. Their CT value ranges are similar, and they have similar imaging characteristics.

[0003] Currently, the clinical identification of kidney stones and vascular calcification mainly relies on doctors' manual interpretation of imaging images. This method has the following problems:

[0004] First, kidney stones and vascular calcifications have similar density characteristics on CTA images, making it difficult to distinguish them manually and prone to misjudgment.

[0005] Second, manual image interpretation is inefficient, especially when a large amount of image data needs to be processed, which places a heavy workload on doctors.

[0006] Third, different doctors have different diagnostic experience and skill levels, resulting in subjective and inconsistent diagnostic results;

[0007] Most existing deep learning-based image segmentation methods employ a single-task learning framework, i.e., segmenting a single target. This approach has the following shortcomings when dealing with the joint identification of kidney stones and vascular calcification:

[0008] First, two independent models need to be trained separately, which increases the complexity of model training and deployment;

[0009] Second, independent single-task models cannot utilize the correlation information between the two lesions, thus limiting segmentation accuracy.

[0010] Third, two independent models may produce conflicting segmentation results for the same region, lacking overall consistency constraints;

[0011] Therefore, a multi-task combined segmentation method for renal stones and vascular calcification in coronary CTA images is needed to improve the accuracy and efficiency of clinical diagnosis. Summary of the Invention

[0012] The purpose of this invention is to provide a multi-task joint segmentation method for kidney stones and vascular calcification in coronary CTA images.

[0013] To achieve the above objectives, the present invention provides the following technical solution: a multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images, comprising a segmentation method, the method comprising the following steps:

[0014] S1. Acquire coronary CTA image data and preprocess the coronary CTA image data to obtain standardized image data;

[0015] S2. Construct a multi-task joint segmentation network, which includes a shared feature encoder, a task-specific decoder, and a task interaction module, wherein the task-specific decoder includes a kidney stone segmentation decoder and a vascular calcification segmentation decoder.

[0016] S3. Input the standardized image data into the shared feature encoder to extract multi-scale feature representations;

[0017] S4. Enhance the multi-scale feature representation by inter-task feature interaction through the task interaction module to obtain the enhanced feature representation;

[0018] S5. Input the enhanced feature representation into the kidney stone segmentation decoder and the vascular calcification segmentation decoder respectively, and output the kidney stone segmentation result and the vascular calcification segmentation result respectively;

[0019] S6. Based on the segmentation results of the kidney stones and the segmentation results of the vascular calcifications, generate the final joint segmentation mask image.

[0020] As a further aspect of the present invention: the preprocessing of the coronary CTA image data in step S1 includes:

[0021] The coronary CTA image data is subjected to grayscale normalization processing to map the pixel values ​​to a preset range;

[0022] The spatial resolution of the normalized image data is adjusted to unify the image size.

[0023] Adjust the window width and window level of the adjusted image data to enhance the contrast of the target area.

[0024] As a further aspect of the present invention: the shared feature encoder adopts a multi-level coding structure, each coding level including a convolutional unit and a downsampling unit, and each coding level sequentially extracts feature representations at different scales.

[0025] As a further aspect of the present invention: the task interaction module includes an attention mechanism submodule and a feature fusion submodule.

[0026] The attention mechanism submodule performs weighted processing on features from different tasks to generate inter-task attention weights.

[0027] The feature fusion submodule adaptively fuses kidney stone-related features and vascular calcification-related features according to the inter-task attention weights.

[0028] As a further aspect of the present invention: the attention mechanism submodule generates inter-task attention weights in the following manner:

[0029] Calculate the characteristic statistics of kidney stone features and vascular calcification features respectively;

[0030] Calculate the correlation measure between features based on the aforementioned feature statistics;

[0031] Adaptive attention weight coefficients are generated based on the relevance metric.

[0032] As a further aspect of the present invention: both the kidney stone segmentation decoder and the vascular calcification segmentation decoder adopt a multi-level decoding structure, each decoding level including an upsampling unit and a feature reconstruction unit, and are fused with the encoder features of the corresponding level through a skip connection method.

[0033] As a further aspect of the present invention: when training the multi-task joint segmentation network, a joint loss function is adopted, which includes a kidney stone segmentation loss term, a blood vessel calcification segmentation loss term, and a task consistency constraint term.

[0034] As a further aspect of the present invention: the task consistency constraint term is used to constrain the mutual exclusion of kidney stone segmentation results and vascular calcification segmentation results in spatial distribution. The mutual exclusion means that at the same pixel position, the sum of the kidney stone segmentation probability and the vascular calcification segmentation probability does not exceed a preset threshold. The preset threshold is 0.8 to 1.2, preferably 1.0.

[0035] As a further aspect of the present invention: the process of generating the final joint segmentation mask image in step S6 includes:

[0036] Post-processing is performed on the segmentation results of the kidney stones and the segmentation results of the vascular calcifications to remove isolated noise points;

[0037] The processed two types of segmentation results are fused with labels to generate a joint segmentation mask image containing labels for kidney stone regions and vascular calcification regions.

[0038] As a further aspect of the present invention, the segmentation method further includes a step of visualizing the segmentation results, wherein the segmentation results of the kidney stones and the segmentation results of the vascular calcifications are superimposed on the original CTA image in different colors.

[0039] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0040] 1. This invention uses a multi-task joint learning framework to simultaneously segment kidney stones and vascular calcification in a single network. Compared with the scheme of training two models independently, this reduces model complexity and computational resource consumption, and improves the feasibility of practical applications.

[0041] 2. This invention extracts a general feature representation by sharing a feature encoder, enabling two tasks to share underlying feature information, thereby improving the efficiency of feature extraction and the generalization ability of feature representation.

[0042] 3. The task interaction module designed in this invention can realize feature interaction and complementarity between two tasks. By utilizing the complementary information of kidney stones and vascular calcification in terms of anatomical location and morphological features, the segmentation accuracy is improved.

[0043] 4. This invention introduces task consistency constraints during the training process to ensure that the segmentation results of kidney stones and vascular calcification do not overlap in space, avoiding conflicting predictions that may be generated by independent models and improving the reliability of the results.

[0044] 5. By jointly optimizing two tasks, this invention enables the network to learn more robust feature representations, thereby improving the model's adaptability to different scanning parameters and individual differences among patients.

[0045] 6. The automated segmentation method provided by this invention can significantly reduce doctors' image reading time, improve diagnostic efficiency, and at the same time reduce the influence of human subjective factors, thereby improving the consistency and accuracy of diagnosis. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0047] Figure 2 This is a schematic diagram of the architecture of the multi-task joint segmentation network of the present invention;

[0048] Figure 3 This is a schematic diagram of the task interaction module of the present invention;

[0049] Figure 4 This is a flowchart illustrating the specific implementation of the method of the present invention. Detailed Implementation

[0050] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0051] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides a multi-task joint segmentation method for renal stones and vascular calcification in coronary CTA images, including the following steps:

[0054] Step S1: Image Data Acquisition and Preprocessing

[0055] First, coronary CTA image data is obtained from the medical imaging system. CTA images are usually stored in DICOM format, which contains pixel data of the image and metadata information such as scan parameters.

[0056] The acquired raw CTA image data undergoes preprocessing, specifically including:

[0057] (1) Gray-scale normalization processing: The pixel values ​​of the CTA image (in Hounsfield units) are normalized. (This refers to normalizing the pixel value to a range of 0 to 1. Since different devices and scanning parameters may cause variations in pixel value ranges, normalization can eliminate these differences and improve the model's generalization ability. Specifically, let the original pixel value be...) The normalized value is Then the normalization formula is:

[0058] ;

[0059] in, Minimum preset The value is usually -1000. This is the preset maximum HU value, usually set to 3000. The value is the normalized pixel value, and its range is [0,1].

[0060] (2) Spatial resolution adjustment: Adjust the image to a uniform spatial size, such as 512×512 pixels. For images with different original sizes, use bilinear interpolation to resample to ensure that the image size input to the network is consistent.

[0061] (3) Window width and level adjustment: Based on the density characteristics of kidney stones and vascular calcification, set appropriate window width and level parameters. The CT values ​​of kidney stones and vascular calcification are usually around 200. Therefore, the window level is set to 400. The window width is 1500. This enhances the contrast between the calcified area and the surrounding soft tissue.

[0062] After the above preprocessing, standardized image data is obtained, which serves as the input for the subsequent segmentation network.

[0063] Step S2: Construct a multi-task joint segmentation network

[0064] like Figure 2 As shown, a multi-task joint segmentation network is constructed. This network adopts an encoder-decoder architecture and includes the following three core components:

[0065] (1) Shared feature encoder: A five-level coding structure is adopted. Each coding level includes two convolutional units and one downsampling unit. The convolutional unit uses a 3×3 convolutional kernel, combined with batch normalization and activation function. The downsampling unit uses 2×2 max pooling operation to halve the feature map size. The number of feature channels in each level is 64, 128, 256, 512 and 1024 respectively. The encoder extracts feature representations from the input image layer by layer from low level to high level. The low-level features contain details and texture information, and the high-level features contain semantic and contextual information.

[0066] (2) Task-specific decoder: includes two parallel decoding branches: kidney stone segmentation decoder and vascular calcification segmentation decoder. Each decoder adopts a five-level structure symmetrical to the encoder. Each decoding level includes an upsampling unit and a feature reconstruction unit. The upsampling unit uses transposed convolution operation to double the size of the feature map. The feature reconstruction unit includes two convolution units to refine the feature representation. The decoder fuses with the features of the corresponding level of the encoder through skip connections to retain more spatial detail information. Finally, each decoder outputs a segmentation probability map of the same size as the input image.

[0067] (3) Task interaction module: such as Figure 3 As shown, the task interaction module is inserted between the highest level of the encoder and the decoder to enable feature interaction between the two tasks. This module includes an attention mechanism submodule and a feature fusion submodule.

[0068] The attention mechanism submodule processes the features from the encoder to generate attention weights between tasks. Specifically, it first performs global average pooling on the features to obtain statistical description vectors of the features, then calculates the correlation between the features through fully connected layers and activation functions, and finally generates the attention weight coefficients for each of the two tasks.

[0069] The feature fusion submodule adaptively fuses the features of the two tasks based on attention weights. For the kidney stone segmentation task, it not only uses its own features but also introduces weighted vascular calcification features, and vice versa. This cross-fusion mechanism enables the network to utilize complementary information between tasks to improve segmentation performance.

[0070] Step S3: Feature Extraction

[0071] The preprocessed standardized image data is input into a shared feature encoder. The image passes through five levels of the encoder, gradually extracting feature representations at different scales. In the first level, shallow features are extracted, mainly including low-level visual information such as edges and textures. As the level deepens, the features become more abstract. By the fifth level, the features contain rich semantic and contextual information.

[0072] Features from each level are saved for subsequent skip connections, and the high-level features output by the encoder are sent to the task interaction module for further processing.

[0073] Step S4: Enhance inter-task feature interaction

[0074] The task interaction module receives high-level features from the encoder and performs feature interaction enhancement processing between tasks.

[0075] First, the attention mechanism submodule calculates the global statistical information of kidney stone features and vascular calcification features respectively. For the input feature F, its size is C×H×W, where C is the number of channels, and H and W are the height and width of the feature map respectively. Through global average pooling operation, a feature vector of size C×1×1 is obtained.

[0076] Then, the feature vectors of the two tasks are concatenated and transformed nonlinearly through a fully connected layer to calculate the correlation measure between the features. This process can capture the relationship between the two tasks in the feature space.

[0077] Next, attention weight coefficients are generated for each task based on the relevance metric. These weight coefficients reflect the importance of the current feature to each task and are normalized to between 0 and 1 using the Sigmoid activation function.

[0078] Finally, the feature fusion submodule performs weighted fusion of the features of the two tasks according to the attention weights. For the kidney stone segmentation task, the enhanced feature is a combination of the original feature and the weighted vascular calcification feature. The same applies to the vascular calcification segmentation task. The fused enhanced feature representation contains both task-specific information and task-shared information, providing a richer feature representation for subsequent decoding.

[0079] Step S5: Task-Specific Decoding and Segmentation

[0080] The enhanced feature representations output by the task interaction module are input into the kidney stone segmentation decoder and the vascular calcification segmentation decoder, respectively.

[0081] During the decoding process, each decoding level first doubles the size of the feature map through an upsampling operation, and then fuses it with the features from the corresponding level of the encoder through skip connections. This skip connection mechanism can combine high-level semantic information and low-level detail information, effectively preserve spatial location information, and improve the precision of segmentation.

[0082] The fused features are further processed by the feature reconstruction unit, which refines the feature representation through multiple convolutional operations. The output of each decoding layer serves as the input to the next layer, restoring the spatial resolution of the feature map layer by layer.

[0083] In the final decoding layer, the feature map is restored to the same size as the input image. It is then passed through a 1×1 convolutional layer and a sigmoid activation function to output the probability value of each pixel belonging to the target category, resulting in the segmentation results for kidney stones and vascular calcification. Both of these segmentation results are probability maps of the same size as the input image, where the value of each pixel represents the probability that the location belongs to the corresponding target.

[0084] Step S6: Generate joint segmentation mask

[0085] The two segmentation results obtained in step S5 are post-processed and fused to generate the final joint segmentation mask image;

[0086] First, the segmentation results of kidney stones and vascular calcifications are binarized. The probability threshold is set to 0.5. Pixels with a probability value greater than 0.5 are marked as foreground (i.e., target area), and pixels with a probability value less than 0.5 are marked as background.

[0087] Then, morphological post-processing is performed on the binarized result, including opening and closing operations, to remove isolated small noise points and fill small holes inside the target region, making the segmentation boundary smoother and more continuous.

[0088] Next, the processed kidney stone segmentation results and vascular calcification segmentation results are fused with labels. The kidney stone region is assigned a label value of 1, the vascular calcification region is assigned a label value of 2, and the background region is assigned a label value of 0. The generated joint segmentation mask image is a single-channel image of the same size as the input image, and the value of each pixel represents the category to which the location belongs.

[0089] Finally, the segmentation results can be visualized, with the kidney stone area displayed in red and the vascular calcification area displayed in blue on the original CTA image, making it easier for doctors to observe the segmentation results and make a diagnosis intuitively.

[0090] Step S7: Network Training

[0091] Before practical application, the multi-task joint segmentation network needs to be trained. The training data includes coronary CTA images with annotated areas of kidney stones and vascular calcification.

[0092] The training process uses a joint loss function, which consists of three parts:

[0093] (1) Kidney stone segmentation loss term: measures the difference between the predicted kidney stone segmentation result and the actual annotation, using the cross-entropy loss function;

[0094] (2) Vascular calcification segmentation loss term: measures the difference between the predicted vascular calcification segmentation result and the actual annotation, and also uses the cross-entropy loss function;

[0095] (3) Task consistency constraint: constrains the spatial exclusivity of the segmentation results of two tasks. Since kidney stones and vascular calcification are two different lesions, they should not overlap anatomically. Therefore, this constraint is introduced to penalize the overlapping area of ​​the two segmentation results. The task consistency constraint measures the degree of overlap by calculating the sum of the pixel-wise product of the two segmentation results. The smaller the value, the better the consistency.

[0096] The expression for the joint loss function L is:

[0097] ;

[0098] in, For the kidney stone segmentation loss term, cross-entropy loss is used:

[0099] ;

[0100] For the vascular calcification segmentation loss term, cross-entropy loss is used:

[0101] ;

[0102] For task consistency constraints:

[0103] ;

[0104] The weighting coefficient for consistency constraints is typically set between 0.1 and 0.5, preferably between 0.2 and 0.3, with an optimal value of 0.25.

[0105] in, Total number of pixels , These are the true labels (0 or 1) for kidney stones and vascular calcification, respectively. , These are the predicted probability values.

[0106] Optimization algorithms such as stochastic gradient descent or adaptive moment estimation are used to update network parameters through backpropagation and minimize the joint loss function. Data augmentation strategies, including random rotation, flipping, and scaling, are employed during training to increase the diversity of training samples and improve the model's generalization ability.

[0107] When the loss on the validation set no longer decreases or reaches the preset number of training epochs, training is stopped, and the parameters of the best-performing model are saved for subsequent inference and prediction.

[0108] Example 2

[0109] This embodiment optimizes and improves the structure of the multi-task joint segmentation network based on Embodiment 1.

[0110] In the shared feature encoder, a residual connection structure is added after each convolutional unit. The residual connection adds the input features directly to the output features through short-circuit connections, which can effectively alleviate the gradient vanishing problem in deep networks and improve the training effect of the network. The relationship between the output Y of the residual unit and the input X is as follows:

[0111] ;

[0112] in, The transformation function representing the convolutional unit includes two 3×3 convolutional layers, batch normalization, and ReLU activation. The input features have dimensions C×H×W. To output features, dimensions and Similarly, addition is an element-wise addition operation.

[0113] In the task interaction module, a multi-head attention mechanism is introduced, which divides the features into multiple subspaces, calculates the attention weights independently in each subspace, and then splices and merges the results of each subspace. The multi-head attention mechanism can capture the relationship between features from different perspectives and improve the expressive power of feature interaction.

[0114] In the task-specific decoder, the skip connection method is improved to the dense connection method. In addition to fusing the features of the corresponding level of the encoder, the features of all levels before the current decoding level are also fused. This dense connection method can make full use of multi-scale feature information and further improve the segmentation accuracy.

[0115] Through the above improvements, the network's feature extraction and representation capabilities are enhanced. In actual testing, the segmentation accuracy is improved by approximately 3 percentage points compared to the basic version.

[0116] Example 3

[0117] This embodiment is optimized for a specific clinical application scenario.

[0118] In the preprocessing stage, a region of interest (ROI) extraction step is added. By analyzing the anatomical structure information of CTA images, the region containing the kidney and major blood vessels is automatically identified and extracted, reducing the processing range and improving computational efficiency. ROI extraction can adopt a threshold-based segmentation method. First, high-density bone structures are identified, and then the kidney and blood vessel regions are located according to their anatomical position.

[0119] In the post-processing stage of segmentation results, three-dimensional connectivity analysis is added. Since CTA images are three-dimensional volume data, performing overall analysis on multiple consecutive slices can improve the coherence of the segmentation results. For small regions that appear isolated on a single slice, if there are corresponding regions in adjacent slices, the region is retained; otherwise, it is removed as noise.

[0120] In the visualization stage, quantitative statistical functions are added to automatically calculate quantitative indicators such as the number, volume, and location distribution of kidney stones and vascular calcifications, and generate structured diagnostic reports to assist doctors in making clinical decisions. For example, the total number of kidney stones and the volume of each stone can be counted, and the Agatston score of vascular calcifications can be calculated to assess the severity of coronary artery calcification.

[0121] The method in this embodiment is closer to the actual clinical workflow and can provide more comprehensive and practical diagnostic information.

[0122] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images, comprising a segmentation method, characterized in that: The method includes the following steps: S1. Acquire coronary CTA image data and preprocess the coronary CTA image data to obtain standardized image data; S2. Construct a multi-task joint segmentation network, which includes a shared feature encoder, a task-specific decoder, and a task interaction module, wherein the task-specific decoder includes a kidney stone segmentation decoder and a vascular calcification segmentation decoder. S3. Input the standardized image data into the shared feature encoder to extract multi-scale feature representations; S4. Enhance the multi-scale feature representation by inter-task feature interaction through the task interaction module to obtain the enhanced feature representation; S5. Input the enhanced feature representation into the kidney stone segmentation decoder and the vascular calcification segmentation decoder respectively, and output the kidney stone segmentation result and the vascular calcification segmentation result respectively; S6. Based on the segmentation results of the kidney stones and the segmentation results of the vascular calcifications, generate the final joint segmentation mask image.

2. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: The preprocessing of the coronary CTA image data in step S1 includes: The coronary CTA image data is subjected to grayscale normalization processing to map the pixel values ​​to a preset range; The spatial resolution of the normalized image data is adjusted to unify the image size. Adjust the window width and window level of the adjusted image data to enhance the contrast of the target area.

3. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: The shared feature encoder adopts a multi-level coding structure, with each coding level including a convolutional unit and a downsampling unit. Each coding level extracts feature representations at different scales in sequence.

4. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: The task interaction module includes an attention mechanism submodule and a feature fusion submodule: The attention mechanism submodule performs weighted processing on features from different tasks to generate inter-task attention weights. The feature fusion submodule adaptively fuses kidney stone-related features and vascular calcification-related features according to the inter-task attention weights.

5. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 4, characterized in that: The attention mechanism submodule generates inter-task attention weights in the following manner: Calculate the characteristic statistics of kidney stone features and vascular calcification features respectively; Calculate the correlation measure between features based on the aforementioned feature statistics; Adaptive attention weight coefficients are generated based on the relevance metric.

6. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: Both the kidney stone segmentation decoder and the vascular calcification segmentation decoder adopt a multi-level decoding structure. Each decoding level includes an upsampling unit and a feature reconstruction unit, and is fused with the encoder features of the corresponding level through a skip connection method.

7. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: When training the multi-task joint segmentation network, a joint loss function is used, which includes a kidney stone segmentation loss term, a blood vessel calcification segmentation loss term, and a task consistency constraint term.

8. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 7, characterized in that: The task consistency constraint is used to constrain the mutual exclusivity of the spatial distribution of kidney stone segmentation results and vascular calcification segmentation results.

9. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: The process of generating the final joint segmentation mask image in step S6 includes: Post-processing is performed on the segmentation results of the kidney stones and the segmentation results of the vascular calcifications to remove isolated noise points; The processed two types of segmentation results are fused with labels to generate a joint segmentation mask image containing labels for kidney stone regions and vascular calcification regions.

10. The multi-task joint segmentation method for renal calculi and vascular calcification in coronary CTA images according to claim 1, characterized in that: The segmentation method also includes a step of visualizing the segmentation results, whereby the segmentation results of the kidney stones and the segmentation results of the vascular calcifications are superimposed on the original CTA image in different colors.

Images