Double-branch fusion heart image segmentation method and system based on memory queue

Abstract

The invention provides a double-branch fusion heart image segmentation method and system based on a memory queue, and relates to the technical field of image segmentation, and the method comprises the steps: inputting a to-be-segmented heart image into CNN-ViT double branches, and generating local features and global features after anatomical prior enhancement; fusing the local features and the global features to obtain image segmentation features; decoding the image segmentation features to obtain a final segmentation result; the anatomical prior enhancement comprises the following steps: clustering and storing historical image features to construct a memory queue, selecting historical image features most similar to a heart image to be segmented from the memory queue, respectively carrying out attention calculation on extracted local features and global features through adjacent window similarity based on the historical image features, and carrying out anatomical prior enhancement on the extracted local features and global features. Dissecting local features and global features after prior enhancement; through CNN-ViT double branches and a memory queue, extraction and enhancement of local features and global features are realized, and segmentation precision and clinical rationality are improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of image segmentation, and in particular to a dual-branch fusion cardiac image segmentation method and system based on a memory queue. Background Art

[0002] In recent years, deep learning technology has significantly promoted the development of medical image segmentation. Convolutional neural networks (CNNs) have become the mainstream solution with their advantages of local perception and translation invariance. U-Net and its improved models have achieved high accuracy in cardiac CT / MRI segmentation through encoder-decoder architecture and skip connections. However, the inductive bias of CNN limits its ability to capture long-distance dependencies, resulting in insufficient global context modeling of the overall anatomical structure of the heart (such as the spatial arrangement of the ventricles and atria, myocardial continuity). For example, in cardiac MRI, due to inter-slice motion artifacts or partial volume effects, relying solely on CNN may produce topological errors (such as ventricular wall rupture) or shape distortion (such as excessive contraction of the ventricular cavity).

[0003] At the same time, the Visual Transformer (ViT) achieves global context modeling through the self-attention mechanism and performs well in natural image tasks. ViT divides images into serialized image patches and captures long-range feature associations through multi-head attention, which is theoretically more suitable for modeling the overall anatomical consistency of the heart organ. However, the application of ViT in medical images faces two major challenges: first, the amount of medical data is limited, and ViT lacks the locality prior inherent in CNN, making it easy to overfit on small data sets. Second, ViT is not sensitive enough to local details (such as myocardial edges and small blood vessel branches) and may lose key anatomical information.

[0004] Based on this, a new cardiac segmentation framework is urgently needed. Through the deep collaboration of CNN and ViT, local features and global context are integrated, while anatomical prior constraints are introduced to improve segmentation accuracy and clinical rationality. However, the following problems currently exist in the collaborative segmentation method between the two: (1) The similarity measurement of medical images needs to meet the matching requirements of local regions and global structures at the same time. However, local features may have semantic conflicts with global anatomical structures, making it difficult for the model to balance the weights of the two.

[0005] (2) The current cardiac segmentation results may output abnormal morphologies that are seriously deviated from the true anatomical structure, such as non-physiological geometric shapes of the ventricular cavity, myocardial wall ruptures, or topological errors in the vascular network. Such problems can easily lead to misdiagnosis in clinical diagnosis.

[0006] (3) Directly calculating the similarities between all image pairs will result in huge computational overhead, especially when processing large-scale image datasets. Each pair of images needs to be compared in detail. The amount of calculation increases exponentially with the increase in the number of images, consuming a lot of time and computing resources. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a dual-branch fusion cardiac image segmentation method and system based on a memory queue. Through the CNN-ViT dual branches and memory queue, the extraction and enhancement of local features and global features are realized, thereby improving the segmentation accuracy and clinical rationality.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions: A dual-branch fusion cardiac image segmentation method based on a memory queue includes: The heart image to be segmented is input into the dual branches of CNN-ViT to generate local features and global features enhanced by anatomical priors; The enhanced local features and global features are integrated to obtain image segmentation features; Decode the image segmentation features to obtain the final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions: A dual-branch fusion cardiac image segmentation system based on memory queue, including: The feature generation module is configured to: input the heart image to be segmented into the CNN-ViT dual branch to generate local features and global features enhanced by anatomical priors; The feature fusion module is configured to: fuse the enhanced local features and global features to obtain image segmentation features; The feature decoding module is configured to: decode the image segmentation features to obtain a final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the memory queue-based dual-branch fusion cardiac image segmentation method is implemented.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the memory queue-based dual-branch fusion cardiac image segmentation method is implemented.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the dual-branch fusion cardiac image segmentation method based on the memory queue.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The dual-branch fusion cardiac image segmentation method and system based on memory queue proposed in this invention significantly improves the clinical practicality and algorithm robustness of medical image segmentation in the following aspects through multi-module collaborative design and cross-scale feature fusion mechanism: 1. Dynamic fusion of local and global features to solve semantic conflict problems: The CNN branch captures detailed features such as myocardial texture and vascular edges through the local perception ability of the convolution kernel, while the ViT branch uses the self-attention mechanism to model the overall anatomical structure of the heart. The fusion module uses an improved Softmax algorithm to calculate the importance weights and non-importance weights of local features in the channel dimension, align the global semantic context in the spatial dimension, and achieve a dynamic balance between local details and global structures.

[0014] 2. Anatomical prior enhancement based on clustering and memory queue: The K-means algorithm is used to cluster the anatomical structure of the input data set, and images of the same type are assigned to the same sub-cohort of the memory cohort. This design enables the model to quickly retrieve similar case features during training and inference, significantly improving the segmentation and generalization capabilities of rare variants. The memory cohort divides the image into overlapping windows and only calculates the cosine similarity of adjacent windows instead of full-image matching, which reduces the computational complexity while retaining anatomical consistency.

[0015] 3. Computational efficiency and real-time optimization: The CNN and ViT branches of the encoder interact with each other after downsampling each layer, avoiding the problem of deep feature misalignment in traditional serial architectures. The fusion module uses sparse matrix compression technology to sparse the global attention matrix of ViT into a block diagonal form, greatly improving the processing speed of cardiac images. The memory queue is divided into categories to reduce the required computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0017] Figure 1 This is the overall structure diagram of the cardiac image segmentation model of Example 1.

[0018] Figure 2 This is the anatomical prior enhancement flowchart of Example 1.

[0019] Figure 3 This is a diagram of the internal structure of the fusion module of Example 1. DETAILED DESCRIPTION

[0020] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0022] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0023] Example 1 In one embodiment of the present disclosure, a dual-branch fusion cardiac image segmentation method based on a memory queue is provided, which combines the CNN-ViT dual branches to achieve the combination of local features and global features, thereby improving the segmentation accuracy and clinical rationality. The specific steps are as follows: Step S1: Input the heart image to be segmented into the dual branches of CNN-ViT to generate local features and global features enhanced by anatomical priors; Step S2: Fusing the enhanced local features and global features to obtain image segmentation features; Step S3: Decode the image segmentation features to obtain the final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement. As an embodiment, the dual-branch fusion cardiac image segmentation method based on the memory queue disclosed in the present invention is divided into two major steps: memory queue construction and cardiac image segmentation, which are described separately below: 1. Memory Queue Construction The memory queue is used to store the characteristic patterns of different anatomical variations. It contains several sub-queues. Each sub-queue stores the image features of the same type of images. Therefore, the construction method is to use the K-means algorithm to cluster the anatomical structure of the historical image data set and assign the image features of the same type of images to the same sub-queue of the memory queue. The specific steps for constructing the memory queue are: (1) Cardiac image clustering: First, traverse the cardiac image files in the specified directory, load all cardiac images, flatten them into one-dimensional vectors and store them in a list. The directory here stores the previously collected historical image datasets.

[0024] Then, K-means (set the number of clusters to N) is used to perform unsupervised clustering on the heart images to generate class labels corresponding to each image.

[0025] Finally, similar images are grouped according to the clustering results. All images in the same group belong to the same category label, which provides a basis for data division for the memory queue in subsequent tasks.

[0026] (2) Constructing a subqueue: Window division and window feature extraction are performed on each image in the same group: the image is divided into In this embodiment, After dividing the window, the feature extractor is used to extract the image features of each window to obtain the window features ,in, , and finally get Window Features .

[0027] All window features obtained from the same group of images are stored in a subqueue, and all subqueues constitute a memory queue.

[0028] In this embodiment, a CNN memory queue for a CNN branch and a ViT memory queue for a ViT branch are respectively constructed by the above method. In the process of constructing the CNN memory queue, the CNN branch in the following cardiac image segmentation model is used as a feature extractor to extract local features as window features to construct a queue; and in the process of constructing the ViT memory queue, the ViT branch in the following cardiac image segmentation model is used as a feature extractor to extract global features as window features to construct a queue.

[0029] 2. Cardiac Image Segmentation The cardiac image segmentation model is used to perform image segmentation on the cardiac image to be segmented, such as Figure 1 As shown, the cardiac image segmentation model includes an encoder, a fusion module and a decoding module, which are described in detail below: 1. Encoder The encoder adopts a CNN-ViT dual branch to generate local features and global features after anatomical prior enhancement. The CNN-ViT dual branch includes a CNN branch and a ViT branch in parallel. The CNN branch includes several consecutive CNN encoders and a CNN memory queue enhancement module, and the ViT branch includes several consecutive ViT encoders and a ViT memory queue enhancement module.

[0030] In the CNN branch, the input heart image is divided into several windows. Here, the division method in the memory queue construction is adopted, that is, it is divided into windows, and perform subsequent feature extraction and feature enhancement in each window, specifically: After the window is divided, the window image enters four consecutive CNN encoders. Each CNN encoder contains a convolutional layer to achieve downsampling, so that the number of channels of the image gradually increases, the image size gradually decreases, and the local features of the image at different scales are extracted. , , is the original input image. The CNN encoder is expressed as:

[0031] in, is convolution, is the stride of the convolution and the size of the convolution kernel.

[0032] Finally, the local features of each window are obtained , the local features of all windows constitute the local features of the heart image ,in For the The local features of a window, namely .

[0033] The local features of the heart image Enter the CNN memory queue enhancement module and perform anatomical prior enhancement by calculating the similarity between each adjacent window.

[0034] In the ViT branch, the input cardiac image is divided into several image block patches. The division method is the same as the division method in the memory queue construction. The patches play the role of windows in the CNN branch. The divided patch images enter four consecutive ViT encoders. Each ViT encoder contains a patch merging module, a transformer module based on self-attention (SA), and a feed-forward neural network (FFN) module to achieve downsampling of cardiac image features and extract global features of different image scales. (i∈(0,1,2,3,4)), is the original input image. The ViT encoder is expressed as:

[0035]

[0036]

[0037]

[0038] in, is a linear layer, is the concatenation of tensors, It is a mask. The specific operation is to extract the features separately. In each 2*2 patch group, patches 0-3 are are query, key, and value, respectively. is the weight matrix of query, key, and value, It is a feed-forward neural network.

[0039] Finally, the global features of each patch are obtained , the global features of all patches constitute the global features of the heart image ,in For the The characteristics of the patch, namely .

[0040] The global features of the heart image Enter the ViT memory queue enhancement module and perform anatomical prior enhancement by calculating the similarity between each adjacent window.

[0041] Both the CNN memory queue enhancement module and the ViT memory queue enhancement module use their own memory queues to perform anatomical prior enhancement on the input local features and global features respectively. The number of subqueues in the memory queue is determined by the number of category labels divided by the cardiac image clustering. Each category of subqueue will only save and update the window features with this category label. The following takes the CNN memory queue enhancement module as an example. Figure 2 As shown, the specific enhancement process is described: (1) Before segmenting the cardiac image, a category prediction is performed on the cardiac image to be segmented, and a corresponding sub-queue is selected from the memory queue based on the predicted category.

[0042] For each input image to be segmented, the image at the head of each sub-memory queue is selected and flattened into one-dimensional vectors. The Euclidean distance between the image to be segmented and the image at the head of each sub-queue is calculated. The category corresponding to the image with the smallest Euclidean distance is the category of the image to be segmented. After feature extraction by the CNN / ViT encoder, the image to be segmented will enter the sub-memory queue corresponding to its category for feature enhancement.

[0043] (2) From the selected sub-queues, use the feature similarity to select the historical image features that are most similar to the input features.

[0044] First, the local features Each window local feature in , and all window local features saved in the same category subqueue Perform cosine similarity calculations one by one and select the most similar window local features , which constitutes the historical image features .

[0045] (3) Based on the selected historical image features , local features are respectively analyzed by neighboring window similarity Perform attention calculation.

[0046] Using the attention mechanism, the features of each window are analyzed through similar historical features of neighboring windows. Perform attention calculation to obtain enhanced window local features , and finally obtain the enhanced local features .

[0047] Specifically, the window For example, taking the current window features For q, the window The 8 adjacent windows are The corresponding window local features in For k and v, attention calculation is performed, and the calculation result is then passed through a feedforward neural network to obtain the anatomical prior enhanced features of the current image.

[0048] In order to ensure the diversity and integrity of the queue image, the window features before enhancement Store it in the memory queue, that is, update the queue as the historical image of the next feature map entering the memory queue. Enhanced local features and global features The fusion will be performed in the fusion module.

[0049] 2. Fusion Module The local features of different scales extracted by the CNN encoder together with the enhanced local features will enter the fusion module and be fused with the global features of the corresponding ViT encoder to obtain image segmentation features of different scales. Therefore, the fusion module includes five fusion units, each of which performs the fusion of features of the corresponding scale. The structure of the fusion unit is as follows: Figure 3 As shown, specifically: (1) Calculation of spatial attention based on local features:

[0050] (2) Calculation of channel attention based on global features:

[0051] (3) Hadamard product operation is performed on local features and global features together:

[0052] in, represents the Hadamard product.

[0053] (4) , and Enter the reSoftmax module together.

[0054] The main structure of the reSoftmax module is the cross attention , the cross attention is similar to the general attention mechanism, the difference is that the reSoftmax function is added to the use of the activation function. The form of the reSoftmax function is as follows:

[0055] in, is an exponential function with base e, is the i-th element of vector z, i∈(1,2,...,N).

[0056] The structure of the reSoftmax module is as follows:

[0057]

[0058]

[0059]

[0060] in, are query, key, and value respectively. is a feed-forward neural network, is the query weight matrix, is the weight matrix of keys and values, for Dimension.

[0061] The common function of Softmax is to highlight the part with larger weight in the attention matrix, which is also the part with higher similarity. ReSoftmax can pay attention to the part with smaller weight at the same time, so that the two feature maps are not missed when fused, achieving better performance.

[0062] 3. Decoding module The decoding module includes four decoders, which are used to decode image segmentation features of different scales to obtain the final segmentation results, specifically: The fusion result of the last layer in the encoder Entering the decoder, the main structure of the decoder is the upsampling module. After each layer of decoder, the image size is expanded and the number of channels is restored. At the same time, the results of the first few fusion units Will correspond to the size Perform residual connection to achieve feature recovery and multi-scale fusion; finally restore the feature image to its original size After segmenting the prediction head, the prediction result is obtained, which is expressed as:

[0063] Finally, the feature image is restored to its original size After splitting the prediction head, the prediction result is obtained.

[0064] The performance of the heart segmentation model is evaluated by some evaluation indicators, including: DICE coefficient, 95% Hausdorff distance (95HD) and mean intersection over union (mIoU). The following are the definitions and calculation methods of these indicators: The DICE coefficient (also known as the F1 score) is an indicator for calculating the similarity between two samples and is often used to evaluate the accuracy of image segmentation. The DICE coefficient calculation formula is: ; in, represents the predicted segmentation result, represents the true segmentation label, is the number of pixels of the intersection of the two, and are the number of pixels of the predicted results and the true labels respectively.

[0065] Mean intersection over union (mIoU) is another important indicator for evaluating model performance in image segmentation tasks. It calculates the ratio of the average intersection over union between the predicted results and the true labels: ; in, is the number of categories, and For the The predicted segmentation results and true labels of the classes.

[0066] Hausdorff distance is an indicator for calculating the maximum deviation between two sets of points. In the field of image segmentation, 95% Hausdorff distance (95HD) is used to evaluate the accuracy of segmentation boundaries, excluding the top 5% of abnormal extreme points in the calculation process and retaining the distance distribution characteristics of the remaining 95% of point pairs. The calculation formula is: ; ; in, Yes To point The Euclidean distance of It means to intercept the first 95% of the value range in the set of minimum distances and take its upper limit as the final result.

[0067] Generate and save segmentation result graphs and display relevant evaluation indicators, which can intuitively show the segmentation effect of the model and help doctors make diagnosis and treatment decisions; at the same time, by providing quantitative evaluation of model performance, it is helpful to further optimize and adjust the model, thereby improving its accuracy and practicality.

[0068] Example 2 In one embodiment of the present disclosure, a dual-branch fusion cardiac image segmentation system based on a memory queue is provided, comprising: The feature generation module is configured to: input the heart image to be segmented into the CNN-ViT dual branch to generate local features and global features enhanced by anatomical priors; The feature fusion module is configured to: fuse the enhanced local features and global features to obtain image segmentation features; The feature decoding module is configured to: decode the image segmentation features to obtain a final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement.

[0069] Example 3 In one embodiment of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the memory queue-based dual-branch fusion cardiac image segmentation method is implemented.

[0070] Example 4 In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the memory queue-based dual-branch fusion cardiac image segmentation method is implemented.

[0071] Example 5 In one embodiment of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the dual-branch fusion cardiac image segmentation method based on the memory queue.

[0072] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0074] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A dual-branch fusion cardiac image segmentation method based on a memory queue, characterized in that: include: The heart image to be segmented is input into the dual branches of CNN-ViT to generate local features and global features enhanced by anatomical priors; The enhanced local features and global features are integrated to obtain image segmentation features; Decode the image segmentation features to obtain the final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement.

2. The dual-branch fusion cardiac image segmentation method based on memory queue according to claim 1, characterized in that: The CNN-ViT dual branch includes a CNN branch and a ViT branch connected in parallel; The CNN branch includes a plurality of consecutive CNN encoders and a CNN memory queue enhancement module, each CNN encoder includes a convolutional layer; The ViT branch includes several consecutive ViT encoders and a ViT memory queue enhancement module, each ViT encoder contains a patch merging module, a self-attention-based transformer module and a feedforward neural network module.

3. The dual-branch fusion cardiac image segmentation method based on memory queue according to claim 1, characterized in that: The clustering and storage of historical image features is to use the K-means algorithm to perform anatomical structure clustering on the historical image data set, and to distribute the image features of the same type of images to the same sub-queue of the memory queue.

4. The dual-branch fusion cardiac image segmentation method based on memory queue according to claim 1, characterized in that: The steps of selecting the historical image features most similar to the heart image to be segmented from the memory queue are as follows: Predict the category of the heart image to be segmented, and select the corresponding sub-queue from the memory queue according to the predicted category; From the selected sub-queues, the historical image features that are most similar to the input features are selected using feature similarity.

5. The dual-branch fusion cardiac image segmentation method based on memory queue according to claim 1, characterized in that: The neighboring window similarity is to divide the image into several windows, and perform attention calculation on the features of the current window through similar historical features of the neighboring windows.

6. The dual-branch fusion cardiac image segmentation method based on memory queue according to claim 1, characterized in that: The fusion adopts an improved Softmax algorithm to calculate the importance weights and non-importance weights of local features in the channel dimension and align the global semantic context in the spatial dimension.

7. A dual-branch fusion cardiac image segmentation system based on a memory queue, characterized in that: include: The feature generation module is configured to: input the heart image to be segmented into the CNN-ViT dual branch to generate local features and global features enhanced by anatomical priors; The feature fusion module is configured to: fuse the enhanced local features and global features to obtain image segmentation features; The feature decoding module is configured to: decode the image segmentation features to obtain a final segmentation result; Among them, the anatomical prior enhancement is to cluster and store historical image features to construct a memory queue, select the historical image features that are most similar to the heart image to be segmented from the memory queue, and based on the historical image features, perform attention calculations on the extracted local features and global features respectively through the similarity of neighboring windows, and the local features and global features after anatomical prior enhancement.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the memory queue-based dual-branch fusion cardiac image segmentation method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, the memory queue-based dual-branch fusion cardiac image segmentation method according to any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the dual-branch fusion cardiac image segmentation method based on the memory queue as described in any one of claims 1-6.

Images