A method, device, equipment and medium for image segmentation of pericardial adipose tissue

Abstract

The application discloses a pericardial fat tissue image segmentation method, device, equipment and medium. The method comprises the following steps: acquiring a cardiac magnetic resonance image of a target object, wherein the cardiac magnetic resonance image contains pericardial fat tissue; segmenting the pericardial fat tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model to obtain target pericardial fat tissue of the target object; the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module; the encoder comprises a double-branch collaborative module; the decoder comprises a multi-layer fusion convolution module; the double-branch collaborative module is used for feature extraction based on a double-branch network; the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions; and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution. The scheme can fully mine semantic information and structural features in the image by using the tissue segmentation model, improve the segmentation performance in a complex scene, and efficiently and accurately segment the pericardial fat tissue.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cardiac medical image segmentation, and in particular to a pericardial fat tissue image segmentation method, device, equipment and medium. BACKGROUND

[0002] With the continuous improvement of the quality of life, people pay more and more attention to their own health status. Studies have shown that pericardial fat tissue, as a fat deposition structure adjacent to the heart and coronary artery, is closely related to the occurrence and development of cardiovascular diseases. Therefore, accurate segmentation of pericardial fat tissue has important clinical significance for early diagnosis, prognosis evaluation and personalized treatment of heart disease.

[0003] In recent years, semantic segmentation technology based on deep learning has made significant progress in the field of medical image analysis, greatly improving the accuracy of medical image segmentation. However, these methods still face challenges when dealing with pericardial fat tissue segmentation tasks in cardiac magnetic resonance images.

[0004] Due to the significant morphological differences of pericardial fat tissue in different individuals, its image features are similar to those of the surrounding tissue, the boundary is blurred, and the noise interference in actual clinical images is significant, making it difficult for general segmentation algorithms to adapt to all situations, which can easily lead to misjudgment or missed detection of the segmentation results, seriously affecting the accuracy and robustness of the segmentation. SUMMARY

[0005] The present application provides a pericardial fat tissue image segmentation method, device, equipment and medium, which can fully exploit the semantic information and structural features in the image, improve the segmentation performance in complex scenarios, and effectively improve the accuracy and robustness of pericardial fat tissue segmentation.

[0006] According to an aspect of the present application, a pericardial fat tissue image segmentation method is provided, the method comprising:

[0007] obtaining a cardiac magnetic resonance image of a target object, the cardiac magnetic resonance image containing pericardial fat tissue;

[0008] segmenting the pericardial fat tissue in the cardiac magnetic resonance image using a pre-trained tissue segmentation model to obtain a target pericardial fat tissue of the target object;

[0009] The tissue segmentation model includes an encoder, a decoder and a spatial channel attention module, the encoder includes a double-branch collaborative module, the decoder includes a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution.

[0010] According to another aspect of the present application, there is provided an image segmentation device for pericardial adipose tissue, the device comprising:

[0011] an image acquisition module configured to acquire a cardiac magnetic resonance image of a target subject, the cardiac magnetic resonance image comprising pericardial adipose tissue;

[0012] an image segmentation module configured to segment the pericardial adipose tissue in the cardiac magnetic resonance image using a pre-trained tissue segmentation model to obtain target pericardial adipose tissue of the target subject.

[0013] wherein the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a dual-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the dual-branch collaborative module is configured to perform feature extraction based on a dual-branch network, the spatial channel attention module is configured to perform feature enhancement in spatial and channel dimensions, and the multi-layer fusion convolution module is configured to perform feature fusion based on grouped convolution.

[0014] According to another aspect of the present application, there is provided an electronic device, the electronic device comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein

[0017] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the image segmentation method for pericardial adipose tissue according to any one of the embodiments of the present application.

[0018] According to another aspect of the present application, there is provided a computer readable storage medium storing computer instructions for causing a processor to implement the image segmentation method for pericardial adipose tissue according to any one of the embodiments of the present application when executed by the processor.

[0019] The technical scheme of the embodiment of the present application firstly acquires a cardiac magnetic resonance image of a target object, and the cardiac magnetic resonance image contains pericardial adipose tissue; then a pre-trained tissue segmentation model is used to segment the pericardial adipose tissue in the cardiac magnetic resonance image, to obtain target pericardial adipose tissue of the target object; wherein the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a double-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution. In this technical scheme, a tissue segmentation model for pericardial adipose tissue is pre-constructed, a double-branch collaborative module is proposed in the encoder part, multi-scale features are collaboratively extracted through the double-branch structure, and the model's ability to capture edge details is enhanced; the spatial channel attention module is used to effectively realize joint enhancement of cross-dimension key information and suppression of background interference, and the model's attention ability to details is enhanced; a lightweight multi-layer fusion convolution module is introduced in the decoder part, efficient decoding and sufficient fusion of deep semantic features are realized; through the tissue segmentation model, semantic information and structural features in the image can be fully mined, the segmentation performance in a complex scene is improved, the accuracy and robustness of pericardial adipose tissue segmentation are effectively improved, and it is of great significance for improving early screening and accurate diagnosis and treatment of cardiovascular diseases.

[0020] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 is a flow chart of a pericardial adipose tissue image segmentation method according to the first embodiment of the present application;

[0023] Figure 2 is a general structure diagram of a tissue segmentation model according to the first embodiment of the present application;

[0024] Figure 3 is a structure schematic diagram of a double-branch collaborative module according to the first embodiment of the present application;

[0025] Figure 4is a structural schematic diagram of a spatial channel attention module according to embodiment one of the present application;

[0026] Figure 5 is a structural schematic diagram of a multi-layer fusion convolution module according to embodiment one of the present application;

[0027] Figure 6 is a flow chart of a pericardial adipose tissue image segmentation method according to embodiment two of the present application;

[0028] Figure 7 is a structural schematic diagram of a pericardial adipose tissue image segmentation device according to embodiment three of the present application;

[0029] Figure 8 is a structural schematic diagram of an electronic device implementing a pericardial adipose tissue image segmentation method according to an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the scope of protection of the present application.

[0031] It should be noted that the terms "first", "second", "target" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0032] Embodiment one

[0033] Figure 1A flowchart of a pericardial fat tissue image segmentation method provided for Embodiment One of the present application. This embodiment can be applied to the efficient and accurate segmentation of pericardial fat tissue in cardiac magnetic resonance images. The method can be performed by a pericardial fat tissue image segmentation device, which can be implemented in the form of hardware and / or software and can be configured in an electronic device with data processing capabilities. As shown in Figure 1 The method comprises the following steps:

[0034] S110, obtaining a cardiac magnetic resonance image of a target object, the cardiac magnetic resonance image containing pericardial fat tissue.

[0035] The cardiac magnetic resonance image can refer to an image generated by imaging the heart using magnetic resonance imaging technology. The target object can refer to the object to which the cardiac magnetic resonance image belongs, i.e., the object from which the cardiac magnetic resonance image is derived.

[0036] In this embodiment, one or more cardiac magnetic resonance images of the target object are first obtained, and it is necessary to ensure that the obtained cardiac magnetic resonance image contains pericardial fat tissue, so that the corresponding pericardial fat tissue can be identified from the cardiac magnetic resonance image in the subsequent process.

[0037] S120, segmenting the pericardial fat tissue in the cardiac magnetic resonance image using a pre-trained tissue segmentation model to obtain target pericardial fat tissue of the target object; wherein the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a double-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution.

[0038] The tissue segmentation model can be a deep learning model that can be used to identify and segment pericardial fat tissue from a cardiac magnetic resonance image. The target pericardial fat tissue can refer to the pericardial fat tissue region identified by the tissue segmentation model.

[0039] In this embodiment, the tissue segmentation model needs to be constructed in advance, and the constructed tissue segmentation model is trained using a full-supervised learning method to obtain the final tissue segmentation model. Specifically, as shown in Figure 2As shown, the tissue segmentation model includes three parts of an encoder, a decoder and a spatial channel attention module. The encoder part introduces a double-branch collaborative module to cooperatively extract multi-scale features through a double-branch structure, thereby enhancing the model's ability to capture edge details. The double-branch network can refer to two different network models that can be used for image feature extraction. For example, the double-branch network can be any two of CNN, U-NET, Transformer, KAN and deep separable convolution network.

[0040] In this embodiment, the optional double-branch collaborative module includes a KAN module and a deep separable convolution residual module. The KAN module includes a KAN linear layer, a deep separable convolution, a batch normalization, a layer normalization and a ReLU activation function. The deep separable convolution residual module includes a deep separable convolution, a point-wise convolution, a batch normalization and a GeLU activation function.

[0041] It should be noted that multiple KAN linear layers can be provided in the KAN module, for example, 3, and a deep separable convolution, a batch normalization and a ReLU activation function are used after each KAN linear layer. Multiple point-wise convolutions can be provided in the deep separable convolution residual module, for example, 2, and a GeLU activation function and a batch normalization are used after each point-wise convolution.

[0042] By such a setting, as shown in Figure 3 As shown, a double-branch encoder structure that integrates the KAN module and the deep separable convolution residual module is designed, which significantly improves the multi-scale feature extraction capability and the morphological adaptability. The KAN network and the deep separable residual convolution network are introduced in the encoder part to construct a double-branch collaborative extraction architecture. The combination of the KAN network and the deep separable convolution residual network makes the two branches cooperate with each other and complement each other in the feature extraction process, so that the model has higher perceptual sensitivity in terms of morphological boundary and spatial position of pericardial fat tissue. The double-branch collaborative extraction architecture realizes the complementary integration of multi-scale semantic features and spatial structure information while maintaining the efficiency of parameters, thereby significantly enhancing the model's representation ability for complex boundaries of pericardial fat tissue. Especially when dealing with anatomical structure variations caused by factors such as body size difference and uneven fat distribution of the target object (such as a patient), the stability and robustness of the model are better than those of the traditional single-branch network.

[0043] As shown in Figure 2As shown, to further strengthen the semantic consistency of feature transmission, a spatial channel attention module is introduced at the skip connection of the encoder-decoder architecture, which can be used for feature enhancement in the spatial and channel dimensions, better capturing the key areas of the image, suppressing irrelevant information interference, and thereby enhancing the model's attention to details. It should be noted that the spatial channel attention module can be set as an integral module, or can be further set as a spatial attention module and a channel attention module, which respectively realize spatial feature enhancement and channel feature enhancement.

[0044] In this embodiment, optionally, the spatial channel attention module includes a spatial attention module and a channel attention module, wherein the spatial attention module is used for spatial feature enhancement, and the channel attention module is used for channel feature enhancement.

[0045] As shown in Figure 4 , the spatial channel attention module includes two substructures: one is a spatial attention module, as shown in Figure 4 (a), which uses different scale convolution kernels (such as 3x3, 5x5, 7x7, 9x9) to extract spatial response features in parallel, for enhancing the positioning ability of the target region; the other is a channel attention module, as shown in Figure 4 (b), which uses a multi-head self-attention mechanism to model the weighting of information between channels, effectively highlighting the response strength of the key channels for pericardial fat tissue segmentation. Finally, the spatial channel attention module jointly enhances the feature map in the spatial and channel dimensions, effectively suppressing background interference and focusing on the region of interest, thereby providing high-quality deep feature expression support for the subsequent decoder.

[0046] Through such a setting, by introducing the spatial channel attention module, the joint enhancement of cross-dimension key information and the suppression of background interference are effectively realized. Compared with the existing pericardial fat tissue segmentation method which only focuses on the spatial or channel dimension, the present application jointly introduces a spatial attention mechanism with a multi-scale convolution kernel and a channel attention mechanism driven by a multi-head self-attention mechanism, and deeply strengthens the intermediate features at the skip connection stage. Among them, the spatial attention module enhances the attention to important features by processing the input feature map in the spatial dimension; the channel attention module focuses on the learning of channel features, and dynamically calibrates the importance of channel features through the multi-head self-attention mechanism. This fusion strategy not only improves the focusing ability of the model on the target region, but also has stronger background suppression ability, significantly improving the segmentation performance of low-contrast or boundary fuzzy regions in the image.

[0047] As shown in Figure 2As shown, the decoder part introduces a multi-layer fusion convolution module, adopts grouped convolution, effectively divides the input features into two groups for processing, and each sub-channel can independently perform convolution operation to effectively extract features at different levels. Not only can it reduce the amount of calculation, but also enhance the diversity of feature fusion, so that the model can extract more rich features at different levels, which helps to better fuse high-dimensional space information.

[0048] In this embodiment, optionally, the multi-layer fusion convolution module includes grouped convolution, point-wise convolution, GeLU activation function and batch normalization, and the grouped convolution is used for feature extraction on the encoder features and the decoder features after upsampling respectively.

[0049] Specifically, as shown in the figure, Figure 5 As shown, the convolution operation is divided into two groups, and feature extraction is performed on the encoder features and the decoder features after upsampling respectively. The size of the convolution kernel of the grouped convolution is 3x3, which can effectively reduce the amount of calculation and improve the feature fusion effect. At the same time, in order to further strengthen the feature fusion, the same point-wise convolution with reverse bottleneck structure as in the encoder is used after the grouped convolution. GeLU activation function and batch normalization layer are used after each convolution layer to enhance the nonlinear expression ability of the model and speed up the training process. Compared with the traditional ReLU activation function, the GeLU activation function is more efficient in processing the nonlinear relationship of input features, which helps the model to capture subtle changes in the image more accurately, thereby improving the sensitivity to details. The multi-layer fusion convolution module effectively integrates the features of the encoder and the decoder at different levels and scales through efficient convolution operation and fine feature fusion design, fully utilizes the advantages of the network in processing complex image tasks, and enables the network to retain key detail information during the process of gradually restoring image resolution, greatly improving the network's processing capability for pericardial fat tissue magnetic resonance images, especially in detail capture and feature fusion.

[0050] By such a setting, the use of a lightweight multi-layer fusion convolution module realizes efficient decoding and full fusion of deep semantic features. The multi-layer fusion convolution module introduced in the decoder divides high-dimensional features into two subgroups by channel, performs parallel convolution processing respectively, and realizes cross-group information interaction and feature reconstruction in the subsequent fusion stage. This design not only effectively reduces the computational redundancy in the decoding process, but also improves the matching efficiency of high-level semantic and low-level detail information, thereby enhancing the boundary clarity and morphological consistency of the segmentation result.

[0051] In the model training stage, the pre-processed data set is fully supervised training. With the pre-labeled pericardial fat tissue label as the supervision target, the weighted combination of boundary weighted cross entropy loss function and Dice loss function is used to enhance the pixel level classification accuracy and regional overlap, so as to effectively optimize the segmentation performance of the model. After training, the trained tissue segmentation model is used to automatically segment the pericardial fat tissue of the unlabeled cardiac magnetic resonance image. Without human intervention, the spatial distribution of the pericardial fat tissue can be quickly located and extracted.

[0052] Specifically, the model training process includes the following steps S1-S3:

[0053] S1, a plurality of cardiac magnetic resonance images containing pericardial fat tissue are obtained to form a data set, and the data set is pre-processed. For example, the present application uses the public MRPEAT data set as the basis for training and evaluation. The data set contains short-axis cardiac magnetic resonance images of 150 subjects in total, covering three types of people: hypertrophic cardiomyopathy, acute myocardial infarction and normal control, 50 cases for each type. Each subject has 20 to 40 groups of dynamic image sequences collected in a complete cardiac cycle, each sequence contains 8 to 18 frames, which can reflect the anatomical and functional status of the heart at the two key stages of diastole and systole. All images are labeled by professional medical imaging personnel to mark the pericardial fat tissue area, the labeling standard is unified, and the label quality is reliable.

[0054] S2, the obtained MRPEAT data set is pre-processed. In order to adapt to the input requirements of the model, the data set needs to be pre-processed, which can specifically include uniform cropping and scaling of image size, and standardization to a fixed resolution under the premise of preserving the integrity of the anatomical structure, and then normalizing the image intensity to reduce the inconsistency of gray scale caused by imaging equipment and individual differences.

[0055] S3, based on the pre-constructed tissue segmentation model, the pre-processed MRPEAT data set is fully supervised training. For example, the present application uses Dice loss and boundary weighted cross entropy loss as training loss to improve the accuracy of the model. The design of Dice loss aims to promote the model to accurately predict the target area by measuring the overlap between the predicted result and the true label. The boundary weighted cross entropy loss adds weight to the boundary area based on the traditional cross entropy loss, thereby strengthening the model's attention to the pericardial fat tissue boundary and ensuring that the model learns the boundary details better.

[0056] The calculation formula of the Dice loss is: . Wherein, is the Dice loss, is the true label, is the model prediction result, is a smoothing factor. The formula of the boundary weighted cross-entropy loss is: wherein, is the boundary weighted cross-entropy loss, is the total number of pixels involved in the calculation; is the weight of the pixel , which is used to adjust the degree of attention of the loss function to the boundary area. The formula of the weight is: wherein, is the distance of the pixel to the nearest boundary, is used to control the degree of weight enhancement of the boundary area, is used to control the decay speed of the weight with the distance. Through this weighting strategy, the loss contribution of the boundary area will be amplified, ensuring that the model can more accurately capture the boundary details of the pericardial fat tissue and improve the overall segmentation performance. The formula of the total loss for training is: wherein, is the parameter for balancing the Dice loss and the boundary weighted cross-entropy loss, is the current training round, is the maximum training round.

[0057] After the model training is completed, the unlabeled cardiac magnetic resonance image is input into the trained tissue segmentation model, and the corresponding pericardial fat tissue prediction mask can be automatically output. In this process, no manual intervention is required, and the automatic segmentation of the pericardial fat tissue area can be efficiently and accurately completed. The prediction results generated by the model show good consistency in terms of structural integrity and regional positioning accuracy, and are expected to provide reliable data support for the diagnosis, risk assessment and treatment planning of related diseases in clinical practice.

[0058] In this embodiment, optionally, the pericardial fat tissue in the cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, including the following steps A1-A5:

[0059] A1, the cardiac magnetic resonance image is data remodeled to obtain a reconstructed image, and the reconstructed image is sequentially processed by a KAN linear layer, a depth separable convolution, a batch normalization and a ReLU activation function to obtain a first processing result.

[0060] In this embodiment, an innovative double-branch collaborative module is designed in the encoder part, which introduces a KAN module and a depth separable convolution residual module as two feature extraction paths respectively, aiming to fuse local fine texture information and global context semantics. Specifically, for the KAN module, as shown in the left branch, Figure 3 the input cardiac magnetic resonance image is first reshaped into a 2D patch sequence , to adapt the processing mode of KAN linear layer. Wherein, , , respectively represent the height, width and channel number of the image. Then it is sent to the continuous stack of three KAN linear layers for processing, and the core design is based on B-spline function, which replaces the traditional weight parameter and helps to capture complex segmentation details. After each KAN linear layer, the input features are further optimized by depth separable convolution, batch normalization and ReLU activation function in turn, and finally the first processing result can be obtained.

[0061] Wherein, step A1 can be represented as: . Wherein, represents the first processing result; represents the depth separable convolution; is batch normalization. In order to ensure that the output of the convolution layer has good balance and stability, reduce the training instability problem caused by the change of input data, and at the same time alleviate the problem of gradient disappearance and improve the convergence speed. By using KAN network, the data can be fitted more accurately, the approximation error is reduced, and when processing the segmentation boundary of pericardial fat tissue, the ability to capture image details is enhanced, and the segmentation precision is improved.

[0062] A2, the reconstructed image and the first processing result are connected by residual connection to obtain the second processing result, and the second processing result is normalized by layer to obtain the first branch result.

[0063] In this embodiment, after obtaining the first processing result, the reconstructed image is added to the first processing as residual by using residual connection to obtain the second processing result, and then the first branch result is obtained by layer normalization as the output result of the double branch collaborative module. Wherein, the processing process of step A2 can be represented as: , wherein, is layer normalization. Using layer normalization can perform more fine-grained normalization within each sample, thereby improving the performance of the model at the microscopic level; is the second processing result, is the first branch result.

[0064] A3, the cardiac magnetic resonance image is sequentially processed by depth separable convolution, GeLU activation function and batch normalization to obtain the third processing result, and the cardiac magnetic resonance image and the third processing result are connected by residual connection to obtain the fourth processing result.

[0065] Specifically, for the depth separable convolution residual module, such as Figure 3As shown in the right branch, the spatial dimension information of the cardiac magnetic resonance image is first extracted using the depth separable convolution, and the original features are reserved using the residual connection. The GeLU activation function and the batch normalization layer are used after each convolution. Among them, step A3 can be represented as: wherein, represents the third processing result, represents the fourth processing result.

[0066] A4, sequentially passing the fourth processing result through the point-wise convolution, the GeLU activation function and the batch normalization processing to obtain the second branch result.

[0067] In order to better fuse the spatial information and the channel information, two point-wise convolutions are applied after the depth separable convolution, and a reverse bottleneck structure is used. The first point-wise convolution expands the number of channels from C to 4C, and the second point-wise convolution compresses the number of channels back to C, that is, the dimension of the intermediate layer between the two point-wise convolutions is expanded by four times, thereby providing a high-dimensional space for feature mixing, avoiding information loss, and enabling the network to more fully capture and fuse the global spatial information obtained through the depth separable convolution. Similarly, the GeLU activation function and the batch normalization layer are used after each convolution. Among them, step A4 is represented as wherein, represents the point-wise convolution, represents the second branch result (i.e., the output result of the depth separable convolution residual module).

[0068] A5, integrating the first branch result and the second branch result, and taking the integrated result as the processing result of the double-branch collaborative module.

[0069] Finally, the output information (i.e., the first branch result and the second branch result) of the two branch processing is added for information integration, realizing complementary advantages while avoiding information loss, thereby obtaining the final output of the double-branch collaborative module. Among them, step A5 can be represented as: wherein, represents the processing result of the double-branch collaborative module.

[0070] In this embodiment, optionally, the pericardial fat tissue in the cardiac magnetic resonance image is segmented using a pre-trained tissue segmentation model, including the following steps B1-B4:

[0071] B1, performing global average pooling on the cardiac magnetic resonance image in the width dimension and the height dimension respectively to obtain a first pooling result and a second pooling result.

[0072] In this embodiment, as Figure 4As shown in (a), a spatial attention module is used to process the input cardiac MRI image in the spatial dimension to enhance the focus on important channel features. Specifically, global average pooling is performed on the input cardiac MRI image in both the width and height dimensions to obtain the first pooling result and the second pooling result, respectively. This process can be found in the formula. and .in, This represents the input cardiac magnetic resonance image. This represents the number of samples processed in parallel during a single forward / backward propagation. and These represent the results of the first pooling and the second pooling, respectively.

[0073] B2. For the first pooling result and the second pooling result, divide them evenly into multiple groups along the channel dimension. Each group is processed by depthwise separable convolution with different kernel sizes to obtain the first convolution result and the second convolution result.

[0074] To achieve multi-scale feature extraction, the results of the first and second pooling can be uniformly divided into multiple groups along the channel dimension. Each group is then processed using depthwise separable convolutions with different kernel sizes, yielding the corresponding first and second convolution results. For example, ... Figure 4 As shown in (a) in the figure, it can be and The data is uniformly divided into four groups along the channel dimension, and processed using depthwise separable convolutions with four different kernel sizes of 3, 5, 7, and 9. For example, the operation process of step B2 can be represented as follows: ,in, This indicates that a one-dimensional depthwise separable convolution is performed on each channel. This represents the result of the first convolution. Indicates the kernel size, small kernel ( Convolution focuses on local detailed features, with a large kernel ( Convolution captures global contextual information; medium-scale convolution ( This establishes cross-regional semantic relationships, allowing the extraction of local and global spatial information at different scales, enabling the model to learn more detailed spatial features. Through this process... Execution and The same processing steps can yield the second convolution result. .

[0075] B3. For the first convolution result and the second convolution result, perform concatenation and group normalization operations in sequence, and use the sigmoid activation function to generate the first spatial attention weight and the second spatial attention weight.

[0076] After obtaining the first convolution result and the second convolution result, each group of results can be spliced and subjected to group normalization operation, as shown in the following formula: wherein, represents splicing, represents group normalization, represents the result after splicing and group normalization operation. Compared with the traditional batch normalization, the group normalization is more stable for small batch training data, and can reduce the consumption of computing resources. By performing group normalization on the spliced result, the model can better normalize the features and improve the training effect. Next, the first spatial attention weight corresponding to the first pooling result and the second spatial attention weight corresponding to the second pooling result are generated through the sigmoid activation function, as shown in the following formula: wherein, and represent the first spatial attention weight and the second spatial attention weight, respectively.

[0077] B4, element-wise multiplication is performed on the cardiac magnetic resonance image, the first spatial attention weight and the second spatial attention weight, and the multiplication result is taken as the processing result of the spatial attention module.

[0078] After obtaining the first spatial attention weight and the second spatial attention weight, the input feature map (i.e. the cardiac magnetic resonance image), the first spatial attention weight and the second spatial attention weight are element-wise multiplied, so as to obtain the processing result of the spatial attention module. Wherein, step B4 can be represented as: wherein, represents the processing result of the spatial attention module.

[0079] Through such a setting, the model can focus more on important spatial regions in the input feature map through the spatial attention module, thereby improving the expressiveness of the model.

[0080] In the present embodiment, optionally, the pericardial fat tissue in the cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, and further comprising steps C1-C3 as follows:

[0081] C1, the processing result of the spatial attention module is subjected to average pooling and group normalization operation in sequence, and the multi-head self-attention of the operation result is determined by using depth separable convolution.

[0082] In the present embodiment, as Figure 4As shown in (b) in the figure, the channel attention module focuses on the learning of channel features, and can dynamically calibrate the importance of channel features through a multi-head self-attention mechanism. Specifically, the processing result of the spatial attention module can be taken as the input of the channel attention module, and first, the spatial dimension is compressed using an average pooling operation, and group normalization is performed, which can be seen from the following formula: wherein, represents the average pooling, represents the result after the average pooling and group normalization operation. Then, a 1x1 depth separable convolution is used to generate a query (Q), key (K), and value (V) matrix and perform multi-head self-attention calculation, which can be seen from the following formula:

[0083] ;

[0084] .

[0085] wherein, represents the query-key-value matrix, represents the multi-head self-attention; is used to calculate the similarity between the query and the key, and is scaled by the vector dimension to avoid large numerical values.

[0086] C2, average pooling is performed on the multi-head self-attention, and a sigmoid activation function is used to generate the channel attention weight.

[0087] Next, the spatial dimension is averaged using average pooling, and a sigmoid activation function is used to generate the channel attention weight, which can be seen from the following formula: . Wherein, represents the channel attention weight.

[0088] C3, the channel attention weight and the processing result of the spatial attention module are multiplied element by element, and the multiplication result is taken as the processing result of the spatial channel attention module.

[0089] Finally, the channel attention weight and the processing result of the spatial attention module are multiplied element by element to obtain the processing result of the spatial channel attention module, which can be seen from the formula .

[0090] By means of the scheme, through the synergistic effect of the two attention modules, the model can perform more accurate weighting and selective attention on the input information at different levels, thereby realizing more detailed feature extraction. The spatial attention module focuses on the local area in the image, and by adaptively adjusting the attention weight of the spatial dimension, the sensitivity of the model to important spatial information is enhanced. The channel attention module focuses on the feature learning at the channel level, and can further improve the feature expression ability by dynamically adjusting the importance of each channel. Especially when dealing with magnetic resonance images such as pericardial fat tissue which has complex morphology and fuzzy boundary, the double attention mechanism can enable the model to automatically focus on important areas in the image, such as the outline of the pericardial fat tissue, thereby effectively improving the segmentation accuracy.

[0091] The technical scheme of the embodiment of the application first acquires a cardiac magnetic resonance image of a target object, and the cardiac magnetic resonance image contains pericardial fat tissue; then a pre-trained tissue segmentation model is used to segment the pericardial fat tissue in the cardiac magnetic resonance image to obtain target pericardial fat tissue of the target object; wherein the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a double-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution. The technical scheme pre-constructs a tissue segmentation model for pericardial fat tissue, proposes a double-branch collaborative module in the encoder part, cooperatively extracts multi-scale features through a double-branch structure, enhances the model's ability to capture edge details, uses a spatial channel attention module to effectively realize joint enhancement of cross-dimension key information and suppression of background interference, enhances the model's attention ability to details, introduces a lightweight multi-layer fusion convolution module in the decoder part, realizes efficient decoding and sufficient fusion of deep semantic features, and through the tissue segmentation model, semantic information and structural features in the image can be fully mined to improve the segmentation performance in complex scenes and effectively improve the accuracy and robustness of pericardial fat tissue segmentation, which is of great significance for improving early screening and accurate diagnosis and treatment of cardiovascular diseases.

[0092] Embodiment two

[0093] Figure 6A flowchart of a pericardial fat tissue image segmentation method provided for the second embodiment of the present application is based on the above-mentioned embodiment for optimization. The specific optimization is that after obtaining the cardiac magnetic resonance image of the target object, the method further includes: pre-processing the cardiac magnetic resonance image, the pre-processing including image size adjustment, image resolution adjustment, and image intensity normalization; correspondingly, the pericardial fat tissue in the cardiac magnetic resonance image is segmented by using the pre-trained tissue segmentation model to obtain the target pericardial fat tissue, which includes: segmenting the pericardial fat tissue in the pre-processed cardiac magnetic resonance image by using the pre-trained tissue segmentation model to obtain the target pericardial fat tissue.

[0094] As shown in Figure 6 , the method of the present embodiment specifically includes the following steps:

[0095] S210, obtaining a cardiac magnetic resonance image of a target object, the cardiac magnetic resonance image containing pericardial fat tissue.

[0096] S220, pre-processing the cardiac magnetic resonance image, the pre-processing including image size adjustment, image resolution adjustment, and image intensity normalization.

[0097] Illustratively, the image size adjustment includes image cropping and scaling. Since the data set is pre-processed during model training to adapt to the model input requirements, in order to improve the accuracy of model prediction, the same data pre-processing as in the model training stage is required for the input cardiac magnetic resonance image in the application stage of the model, specifically including uniform cropping and scaling of the image size, standardizing to a fixed resolution under the premise of preserving the integrity of the anatomical structure, and normalizing the image intensity, thereby reducing the gray inconsistency caused by the imaging equipment and individual differences.

[0098] S230, segmenting the pericardial fat tissue in the pre-processed cardiac magnetic resonance image by using the pre-trained tissue segmentation model to obtain the target pericardial fat tissue.

[0099] Among them, the tissue segmentation model includes an encoder, a decoder, and a spatial channel attention module, the encoder includes a double-branch collaborative module, the decoder includes a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution. The specific implementation process of this step can be referred to the related description of the above-mentioned embodiment, which will not be described here.

[0100] The technical scheme of the embodiment of the present application firstly acquires a cardiac magnetic resonance image of a target object, and the cardiac magnetic resonance image contains pericardial adipose tissue; then, the cardiac magnetic resonance image is preprocessed, and the preprocessing includes image size adjustment, image resolution adjustment and image intensity normalization; finally, the pericardial adipose tissue in the cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model to obtain target pericardial adipose tissue of the target object; wherein, the tissue segmentation model includes an encoder, a decoder and a spatial channel attention module, the encoder includes a double-branch collaborative module, the decoder includes a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution. The technical scheme adds a data preprocessing operation before the pericardial adipose tissue in the cardiac magnetic resonance image is segmented by using the tissue segmentation model, can better adapt to the model input requirements, reduces the gray inconsistency caused by the imaging equipment and individual differences, and is helpful to further improve the prediction accuracy of the tissue segmentation model.

[0101] Embodiment three

[0102] Figure 7 A structure schematic diagram of a pericardial adipose tissue image segmentation device provided by the third embodiment of the present application is shown in the figure. The device can execute the pericardial adipose tissue image segmentation method provided by any embodiment of the present application, has the corresponding function modules and beneficial effects of the execution method. As shown in the figure, the device includes: Figure 7

[0103] The image acquisition module 310 is configured to acquire a cardiac magnetic resonance image of a target object, and the cardiac magnetic resonance image contains pericardial adipose tissue.

[0104] The image segmentation module 320 is configured to segment the pericardial adipose tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model to obtain target pericardial adipose tissue of the target object.

[0105] The tissue segmentation model includes an encoder, a decoder and a spatial channel attention module, the encoder includes a double-branch collaborative module, the decoder includes a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution.

[0106] ​Optionally, the double-branch collaborative module comprises a KAN module and a depth separable convolution residual module, the KAN module comprises a KAN linear layer, a depth separable convolution, a batch normalization, a layer normalization and a ReLU activation function, and the depth separable convolution residual module comprises a depth separable convolution, a point-wise convolution, a batch normalization and a GeLU activation function.

[0107] Optionally, the image segmentation module 320 is configured to:

[0108] perform data remodeling on the cardiac magnetic resonance image to obtain a reconstructed image, and sequentially perform processing on the reconstructed image by a KAN linear layer, a depth separable convolution, a batch normalization and a ReLU activation function to obtain a first processing result;

[0109] perform residual connection on the reconstructed image and the first processing result to obtain a second processing result, and perform layer normalization on the second processing result to obtain a first branch result;

[0110] sequentially perform processing on the cardiac magnetic resonance image by a depth separable convolution, a GeLU activation function and a batch normalization to obtain a third processing result, and perform residual connection on the cardiac magnetic resonance image and the third processing result to obtain a fourth processing result;

[0111] sequentially perform processing on the fourth processing result by a point-wise convolution, a GeLU activation function and a batch normalization to obtain a second branch result;

[0112] integrate the first branch result and the second branch result, and take an integrated result as a processing result of the double-branch collaborative module.

[0113] Optionally, the spatial channel attention module comprises a spatial attention module, and the spatial attention module is configured to perform spatial feature enhancement.

[0114] Correspondingly, the image segmentation module 320 is configured to:

[0115] perform global average pooling on the cardiac magnetic resonance image in a width dimension and a height dimension respectively to obtain a first pooling result and a second pooling result;

[0116] for the first pooling result and the second pooling result, respectively, uniformly divide into a plurality of groups along a channel dimension, and each group is processed by a depth separable convolution with different kernel sizes to obtain a first convolution result and a second convolution result;

[0117] for the first convolution result and the second convolution result, respectively, sequentially perform splicing and group normalization operations, and generate a first spatial attention weight and a second spatial attention weight by using a sigmoid activation function;

[0118] The cardiac magnetic resonance image, the first spatial attention weight and the second spatial attention weight are multiplied element by element, and the multiplication result is taken as the processing result of the spatial attention module.

[0119] Optionally, the spatial channel attention module further comprises a channel attention module, and the channel attention module is used for channel feature enhancement.

[0120] Correspondingly, the image segmentation module 320 is further configured to:

[0121] The processing result of the spatial attention module is sequentially subjected to average pooling and group normalization operations, and a multi-head self-attention of the operation result is determined by using a depth separable convolution;

[0122] The multi-head self-attention is subjected to average pooling, and a channel attention weight is generated by using a sigmoid activation function;

[0123] The channel attention weight and the processing result of the spatial attention module are multiplied element by element, and the multiplication result is taken as the processing result of the spatial channel attention module.

[0124] Optionally, the multi-layer fusion convolution module comprises a grouped convolution, a point-wise convolution, a GeLU activation function and a batch normalization, and the grouped convolution is used for feature extraction on the encoder feature and the decoder feature subjected to up-sampling respectively.

[0125] Optionally, the device further comprises an image preprocessing module configured to:

[0126] After obtaining the cardiac magnetic resonance image of the target object, the cardiac magnetic resonance image is preprocessed, and the preprocessing comprises image size adjustment, image resolution adjustment and image intensity normalization.

[0127] Correspondingly, the image segmentation module 320 is configured to:

[0128] The pericardial fat tissue in the preprocessed cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, and the target pericardial fat tissue is obtained.

[0129] The image segmentation device for pericardial fat tissue provided in the embodiment of the application can execute the image segmentation method for pericardial fat tissue provided in any embodiment of the application, and has the corresponding function modules and beneficial effects of the execution method.

[0130] Embodiment four

[0131] Figure 8A structural diagram of an electronic device 10 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices (e.g., headsets, glasses, watches, etc.), and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application described and / or claimed in this document.

[0132] As shown in Figure 8 The electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., communicatively connected to the at least one processor 11, where the memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer programs stored in the read-only memory (ROM) 12 or loaded into the random access memory (RAM) 13 from the storage unit 18. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0133] Various components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc., an output unit 17, such as various types of displays, speakers, etc., a storage unit 18, such as a magnetic disk, an optical disk, etc., and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0134] The processor 11 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as the image segmentation method of pericardial adipose tissue.

[0135] In some embodiments, the image segmentation method of pericardial adipose tissue can be implemented as a computer program tangibly embodied in a computer readable storage medium, e.g., storage unit 18. In some embodiments, parts or all of the computer program can be loaded and / or installed onto electronic device 10 via, e.g., ROM 12 and / or communication unit 19. When the computer program is loaded onto RAM 13 and executed by processor 11, one or more steps of the image segmentation method of pericardial adipose tissue as described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the image segmentation method of pericardial adipose tissue by way of other any suitable means, e.g., by way of firmware.

[0136] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0137] Computer programs used to implement the processes of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program, when executed, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0138] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0139] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0140] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0141] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0142] It should be understood that the various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0143] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of image segmentation of pericardial adipose tissue, characterized by, The method comprises: acquiring a cardiac magnetic resonance image of a target object, the cardiac magnetic resonance image containing pericardial adipose tissue; segmenting the pericardial adipose tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model to obtain target pericardial adipose tissue of the target object; wherein the tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a double-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the double-branch collaborative module is used for feature extraction based on a double-branch network, the spatial channel attention module is used for feature enhancement in the spatial and channel dimensions, and the multi-layer fusion convolution module is used for feature fusion based on grouped convolution; the double-branch collaborative module comprises a KAN module and a depth separable convolution residual module, the KAN module comprises a KAN linear layer, a depth separable convolution, a batch normalization, a layer normalization and a ReLU activation function, and the depth separable convolution residual module comprises a depth separable convolution, a point-wise convolution, a batch normalization and a GeLU activation function; and the multi-layer fusion convolution module comprises a grouped convolution, a point-wise convolution, a GeLU activation function and a batch normalization, and the grouped convolution is used for feature extraction on the encoder features and the decoder features after up-sampling respectively.

2. The method of claim 1, wherein, segmenting the pericardial adipose tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model comprises: performing data remodeling on the cardiac magnetic resonance image to obtain a reconstructed image, and sequentially processing the reconstructed image through a KAN linear layer, a depth separable convolution, a batch normalization and a ReLU activation function to obtain a first processing result; performing residual connection on the reconstructed image and the first processing result to obtain a second processing result, and performing layer normalization on the second processing result to obtain a first branch result; sequentially processing the cardiac magnetic resonance image through a depth separable convolution, a GeLU activation function and a batch normalization to obtain a third processing result, and performing residual connection on the cardiac magnetic resonance image and the third processing result to obtain a fourth processing result; sequentially processing the fourth processing result through a point-wise convolution, a GeLU activation function and a batch normalization to obtain a second branch result; integrating the first branch result and the second branch result, and taking the integrated result as a processing result of the double-branch collaborative module.

3. The method of claim 1, wherein, The spatial channel attention module comprises a spatial attention module for spatial feature enhancement; correspondingly, segmenting the pericardial adipose tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model comprises: performing global average pooling on the cardiac magnetic resonance image in the width dimension and the height dimension respectively to obtain a first pooling result and a second pooling result; for the first pooling result and the second pooling result, dividing each along the channel dimension into a plurality of groups uniformly, processing each group by using a depth separable convolution with a different kernel size to obtain a first convolution result and a second convolution result; The first spatial attention weight and the second spatial attention weight are generated by using a sigmoid activation function after performing a splicing operation and a group normalization operation on the first convolution result and the second convolution result in sequence respectively. The cardiac magnetic resonance image, the first spatial attention weight and the second spatial attention weight are multiplied element by element, and a multiplication result is taken as a processing result of the spatial attention module.

4. The method of claim 3, wherein, The spatial channel attention module further comprises a channel attention module, and the channel attention module is used for channel feature enhancement. Correspondingly, the pericardial fat tissue in the cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, and the method further comprises the following steps: The processing result of the spatial attention module is subjected to an average pooling operation and a group normalization operation in sequence, and a multi-head self-attention operation result is determined by using a depth separable convolution. The multi-head self-attention is subjected to an average pooling operation, and a channel attention weight is generated by using a sigmoid activation function. The channel attention weight and the processing result of the spatial attention module are multiplied element by element, and a multiplication result is taken as a processing result of the spatial channel attention module.

5. The method according to any one of claims 1 to 4, characterized in that, After obtaining the cardiac magnetic resonance image of the target object, the method further comprises the following steps: The cardiac magnetic resonance image is preprocessed, and the preprocessing comprises image size adjustment, image resolution adjustment and image intensity normalization. Correspondingly, the pericardial fat tissue in the cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, and the method further comprises the following steps: The pericardial fat tissue in the preprocessed cardiac magnetic resonance image is segmented by using a pre-trained tissue segmentation model, and a target pericardial fat tissue is obtained.

6. An apparatus for segmenting an image of pericardial adipose tissue, characterized by comprising: The device comprises: An image acquisition module is configured to acquire a cardiac magnetic resonance image of a target object, and the cardiac magnetic resonance image comprises pericardial fat tissue. An image segmentation module is configured to segment the pericardial fat tissue in the cardiac magnetic resonance image by using a pre-trained tissue segmentation model, and obtain a target pericardial fat tissue of the target object. The tissue segmentation model comprises an encoder, a decoder and a spatial channel attention module, the encoder comprises a double-branch collaborative module, the decoder comprises a multi-layer fusion convolution module, the double-branch collaborative module is configured to perform feature extraction based on a double-branch network, the spatial channel attention module is configured to perform feature enhancement in a spatial and channel dimension, and the multi-layer fusion convolution module is configured to perform feature fusion based on a group convolution; the double-branch collaborative module comprises a KAN module and a depth separable convolution residual module, the KAN module comprises a KAN linear layer, a depth separable convolution, a batch normalization, a layer normalization and a ReLU activation function, and the depth separable convolution residual module comprises a depth separable convolution, a point-wise convolution, a batch normalization and a GeLU activation function; the multi-layer fusion convolution module comprises a group convolution, a point-wise convolution, a GeLU activation function and a batch normalization, and the group convolution is configured to perform feature extraction on the encoder feature and the up-sampled decoder feature respectively.

7. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the image segmentation method of pericardial fat tissue according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to implement the image segmentation method of pericardial fat tissue according to any one of claims 1-5 when executed.

Images