An Abnormal Brain Image Segmentation Method and System Based on Dual Attention Mechanism

Abstract

The present invention provides an abnormal brain image segmentation method and system based on a dual attention mechanism, which relates to the technical field of medical image processing. The abnormal brain image segmentation method includes: acquiring an original abnormal brain image; performing normalization processing on the original abnormal brain image, extracting a texture feature image by using a gray-level co-occurrence matrix, building a convolutional neural network model, inputting the original abnormal brain image and the texture feature image into the built convolutional neural network model for model training, and segmenting a nuclear magnetic resonance brain image based on the trained convolutional neural network model. Among them, a position attention mechanism module and a channel attention mechanism module are introduced into the convolutional neural network model. By combining the gray-level co-occurrence matrix and the dual attention mechanism, the brain healthy tissue and the tumor region can be effectively segmented, and the segmentation accuracy of the three-dimensional abnormal brain image can be improved.

Description

Technical Field

[0001] This application belongs to the technical field of medical image processing, and particularly relates to an abnormal brain image segmentation method and system based on a dual attention mechanism. Background Technique

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Medical image segmentation is a key technology in image analysis and processing. It separates relevant tissues of interest according to the similarity and specificity of regions within the image and extracts relevant features. This is of great significance for the clinical diagnosis and treatment process and is the main premise for all subsequent work. Moreover, the quality of the segmentation effect will directly affect the smooth progress of subsequent information processing work.

[0004] Brain tumors, as abnormal tissues in the brain, are considered a common neurological disease that harms the healthy tissues of the brain, grows rapidly, and is not easily detected. Magnetic resonance imaging (MRI) is a typical non-invasive imaging technology that can produce high-quality brain images without damage and skull artifacts and is considered the main technical means for diagnosing and treating brain tumors. Through multi-modal brain images, doctors can perform quantitative analysis on brain tumors, thereby making the best diagnosis and treatment plan for patients. However, manually segmenting brain tumors from MR images requires a large amount of time, resources, and professional knowledge. In addition, manual segmentation depends on clinical experience, and there are significant differences in the segmentation of each tumor region by different experts. Computer automatic segmentation not only saves time and cost but also improves the objectivity of quantitative analysis. Therefore, automatic segmentation of abnormal brain magnetic resonance images can provide a reliable basis for clinical diagnosis and treatment and pathological research, assisting doctors in making more accurate diagnoses.

[0005] In the early stage of image segmentation, deep learning had not emerged, and traditional machine learning methods were usually used for segmentation. Joseph et al. used thresholds to determine lung tissue, thus achieving automatic segmentation applied to x-ray computed tomography. The threshold-based segmentation method directly utilizes the gray-scale characteristics of the image and has a relatively fast calculation speed, but it can only be applied to images with a large difference between the target and the background. Tang et al. adopted edge segmentation and then region segmentation based on the connectivity of brain tissue structures to display finer brain tissue structures. Juang et al. proposed a color conversion segmentation algorithm based on k-means. The core idea is to convert the input gray-scale image into a color space image, and then perform operations on the image marked with clustering metrics to segment the given MRI image.

[0006] With the continuous breakthrough of deep learning technology, automatic image segmentation methods based on deep learning have also made great progress. Ronneberger et al. proposed a symmetric fully convolutional network U-Net for medical image segmentation. U-Net consists of a contracting path (for downsampling the input image) containing multiple convolutional layers, an expanding path (for upsampling the deep feature maps), and a skip connection (for merging the cropped feature maps from the encoder-decoder network), which greatly improves the segmentation performance of medical images. However, the inventors found that multiple convolutional operations cause the edge information of the feature maps to disappear, affecting the segmentation accuracy of abnormal brain images. Summary of the Invention

[0007] To overcome the deficiencies of the prior art and address problems such as the small number of brain tumor image datasets and the low accuracy of existing segmentation algorithms, this application provides an abnormal brain image segmentation method and system based on a dual attention mechanism.

[0008] The technical solutions adopted in this application are as follows:

[0009] In a first aspect, an embodiment of the present invention provides an abnormal brain image segmentation method based on a dual attention mechanism, including:

[0010] Obtain the original abnormal brain image;

[0011] Perform normalization processing on the original abnormal brain image, generate a texture feature image using the gray-level co-occurrence matrix, build a convolutional neural network model, input the original abnormal brain image and the texture feature image into the built convolutional neural network model for model training, and segment the magnetic resonance brain image based on the trained convolutional neural network model.

[0012] Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit; in the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is incorporated in subsequent upsamplings; the original abnormal brain image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connection.

[0013] In a possible implementation, it further includes: dividing the obtained magnetic resonance brain images into a training set and a validation set according to a set ratio; the magnetic resonance brain images in the training set are used to train the convolutional neural network model, and the magnetic resonance brain images in the validation set are used to test and evaluate the segmentation effect of the convolutional neural network model after the convolutional neural network model is trained.

[0014] In a possible implementation, the statistical mean, standard deviation, contrast, and dissimilarity are calculated using the gray-level co-occurrence matrix to generate texture feature images of the mean, standard deviation, contrast, and dissimilarity.

[0015] In a possible implementation, the encoder includes five convolutional modules. The first convolutional module takes the abnormal brain original image and the generated texture feature image as inputs and performs 3×3×3 convolution. After passing through the batch normalization layer and the LeakyRelu activation, it is short-connected and input into the next convolutional module. The second to fifth convolutional modules use residual blocks to replace the pooling layer.

[0016] In a possible implementation, the residual module includes two units, and each unit consists of regularization, an activation function, and a 3×3×3 convolutional layer.

[0017] In a possible implementation, the decoder includes four upsampling layers. After the first upsampling and concatenation of the feature map, it passes through the concatenated channel attention mechanism module and position attention mechanism module and is input into the residual module. The input of the second upsampling layer includes the output of the previous upsampling layer and the outputs of the depth convolution path and standard convolution path of the abnormal brain original image and texture feature image in the third convolutional layer of the encoder. After the second upsampling and concatenation, it passes through the channel attention mechanism module and is input into the residual module. The third and fourth upsampling layers are in the same way as the second upsampling layer, and finally, the softmax function is used to activate the output of the upsampling stage.

[0018] In a possible implementation, the loss function of the convolutional neural network model is selected as a combination of cross-entropy and dice loss.

[0019] In a second aspect, an abnormal brain image segmentation system based on a dual attention mechanism provided by an embodiment of the present application includes:

[0020] An acquisition module, configured to acquire an abnormal brain original image;

[0021] A segmentation module, configured to perform normalization processing on the abnormal brain original image, extract a texture feature image using the gray-level co-occurrence matrix, build a convolutional neural network model, input the abnormal brain original image and the texture feature image into the built convolutional neural network model for model training, and segment the nuclear magnetic resonance brain image based on the trained convolutional neural network model;

[0022] Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit. In the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and the subsequent upsamplings only incorporate the channel attention mechanism module. The abnormal brain original image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connections.

[0023] In a third aspect, an embodiment of the present invention provides a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the abnormal brain image segmentation method described in the first aspect and any possible implementation manner of the first aspect are executed.

[0024] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is run by a processor, the steps of the abnormal brain image segmentation method described in the first aspect and any possible implementation manner of the first aspect are executed.

[0025] The beneficial effects of this application are as follows:

[0026] 1. Aiming at the problems of few brain tumor image data sets and low accuracy of existing segmentation algorithms, this application provides an abnormal brain image segmentation method based on a dual attention mechanism. The gray-level co-occurrence matrix is used to extract the texture feature image, and the abnormal brain original image and the texture feature image are input into the convolutional neural network model for model training, which can increase the sample data set and is conducive to better extracting image features. At the same time, on the basis of the 3D U-Net network model, a dual attention mechanism is introduced to solve the problem that the edge information of the feature map disappears due to multiple convolutional operations, and improve the segmentation accuracy of abnormal brain images.

[0027] 2. Compared with the existing method for medical image segmentation based on the 3D U-Net network, the average Dice evaluations of the abnormal brain segmentation method provided in this application for tumors, cerebrospinal fluid, white matter, and gray matter of the brain can reach 0.81, 0.91, 0.91, and 0.93 respectively, with higher accuracy. Description of the Drawings

[0028] The accompanying drawings forming a part of this invention are used to provide a further understanding of the invention. The schematic embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0029] Figure 1It is one of the flowcharts of the abnormal brain image segmentation method provided by the embodiments of the present application;

[0030] Figure 2 It is the second flowchart of the abnormal brain image segmentation method provided by the embodiments of the present application;

[0031] Figure 3 It is the texture feature image generated after calculating the mean value, standard deviation, contrast and dissimilarity of the T1-modal image through the gray-level co-occurrence matrix provided by the embodiments of the present application;

[0032] Figure 4 It is the structural schematic diagram of the convolutional neural network model provided by the embodiments of the present application;

[0033] Figure 5 It is the structural schematic diagram of the channel attention mechanism module provided by the embodiments of the present application;

[0034] Figure 6 It is the structural schematic diagram of the position attention mechanism module provided by the embodiments of the present application;

[0035] Figure 7 It is the segmentation result diagram of different convolutional network models provided by the embodiments of the present application;

[0036] Figure 8 It is the structural schematic diagram of the abnormal brain image segmentation system provided by the embodiments of the present application;

[0037] Figure 9 It is the schematic diagram of a computer device provided by the embodiments of the present application. Specific embodiments

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed description is exemplary and is intended to provide further description of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments of the present invention. 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, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Embodiment 1

[0042] Please refer to Figure 1 andFigure 2 , in view of the problems of few brain tumor image datasets and low accuracy of existing segmentation algorithms, this application provides an abnormal brain image segmentation method based on a dual attention mechanism. The device for executing the abnormal brain image segmentation method can be a cloud platform or a server that interacts with the user terminal. Below, from the perspective of the server as the execution subject, the abnormal brain image segmentation method provided by the embodiments of this application will be described. As Figure 1 shown in

[0043] S101: Obtain the original abnormal brain image.

[0044] In a specific implementation, magnetic resonance imaging can generate high-quality brain images without damage and skull artifacts. Therefore, in this embodiment, 3D brain images in the T1 modality are used, and the size of each image is 192×192×160.

[0045] S102: Perform normalization processing on the original abnormal brain image, generate a texture feature image using the gray-level co-occurrence matrix, build a convolutional neural network model, input the original abnormal brain image and the texture feature image into the built convolutional neural network model for model training, and segment the nuclear magnetic resonance brain image based on the trained convolutional neural network model;

[0046] Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit; in the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is incorporated in the subsequent upsamplings; the original abnormal brain image and the texture feature image are respectively subjected to feature extraction through the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connections.

[0047] In a specific implementation, in this embodiment, the acquired abnormal brain original image is first normalized, and then a texture feature image is generated using a gray-level co-occurrence matrix. A convolutional neural network model is built, and the abnormal brain original image and the texture feature image are input into the built convolutional neural network model for model training. Based on the trained convolutional neural network model, magnetic resonance brain images are segmented. In this way, the sample data set can be increased, which is beneficial to better extracting image features. At the same time, the built convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit. In the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is incorporated in the subsequent upsamplings. The abnormal brain original image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connections. In this way, the problem of the disappearance of edge information of the feature map caused by multiple convolutional operations can be solved, and the segmentation accuracy of abnormal brain images can be improved.

[0048] In an embodiment of the present application, as an alternative implementation, the abnormal brain image segmentation method further includes: dividing the acquired magnetic resonance brain images into a training set and a validation set according to a set ratio; the magnetic resonance brain images in the training set are used to train the convolutional neural network model, and the magnetic resonance brain images in the validation set are used to test and evaluate the segmentation effect of the convolutional neural network model after the convolutional neural network model is trained.

[0049] In a specific implementation, the acquired 3D magnetic resonance brain images are divided into a training set and a validation set according to a ratio of 8:2. The convolutional neural network model is trained with the magnetic resonance brain images in the training set. After the convolutional neural network model is trained, the segmentation effect of the convolutional neural network model is tested and evaluated with the magnetic resonance brain images in the validation set, specifically as Figure 2 shown.

[0050] In an embodiment of the present application, as an alternative implementation, the statistical mean, standard deviation, contrast, and dissimilarity are calculated using a gray-level co-occurrence matrix to generate texture feature images of the mean, standard deviation, contrast, and dissimilarity. Among them, the mean is used to reflect the average situation of the gray-level distribution of the image; the standard deviation is used to reflect the situation of the gray-level transformation of the image; the contrast is used to describe the clarity of the image and the depth of the texture; the dissimilarity is used to describe the local changes in the image, and then texture feature maps of the mean, standard deviation, contrast, and dissimilarity are generated, as Figure 3 shown.

[0051] In an embodiment of the present application, as an alternative embodiment, the encoder includes five convolutional modules. The first convolutional module takes the abnormal brain original image and the generated texture feature image as inputs, performs 3×3×3 convolution, and after passing through the batch normalization layer and the LeakyRelu activation, is short-connected and input into the next convolutional module. The second to fifth convolutional modules use residual blocks to replace the pooling layer. Optionally, the residual module includes two units, and each unit consists of regularization, an activation function, and a 3×3×3 convolutional layer.

[0052] As an alternative embodiment, the decoder includes four upsampling layers. After the first upsampling and concatenation of the feature map, it is input into the residual module through the concatenated channel attention mechanism module and position attention mechanism module. The input of the second upsampling layer includes the output of the first upsampling layer and the outputs of the depth convolution path and the standard convolution path of the abnormal brain original image and the texture feature image in the third convolutional layer of the encoder. After the second upsampling and concatenation, it is input into the residual module through the channel attention mechanism module. The third and fourth upsampling layers are in the same manner as the second upsampling layer. Finally, the softmax function is used to activate the output of the upsampling stage.

[0053] In a specific implementation, as Figure 4 shown, the embodiment of the present application is based on 3D U-Net and includes a downsampling stage, an upsampling stage, and an activation layer.

[0054] The downsampling stage includes 5 convolutional modules. The first convolutional module takes the abnormal brain original image and the generated texture feature image as inputs, performs 3×3×3 convolution, and after passing through the batch normalization layer and the LeakyRelu activation, is short-connected and input into the next convolutional module. The 2nd, 3rd, 4th, and 5th convolutional modules use residual blocks to replace the pooling layer. The residual block includes two identical units, and each unit consists of regularization, the LeakyRelu activation function, and a 3×3×3 convolutional layer. The stride of the previous unit is 2, and the stride of the latter unit is 1. For each downsampling operation, the size of the feature map is halved, but the corresponding number of channels is doubled.

[0055] The size of the feature map is restored to the size of the input slice through the upsampling stage. Among them, the depth convolution path and the standard convolution path in the downsampling stage are spliced together through trainable coefficients to obtain a new feature map. The input of the first upsampling layer includes the new feature map and the outputs of the abnormal brain original image and the texture feature image paths in the 4th convolutional layer of the downsampling stage. After the first upsampling and concatenation, it is input into the residual module through the concatenated channel attention mechanism module and position attention mechanism module.

[0056] The input of the second upsampling layer includes the output of the previous upsampling layer and the outputs of the depth convolution path and the standard convolution path of the abnormal brain original image and the texture feature image in the third convolutional layer in the encoder. After the second upsampling and concatenation, it is input into the residual module through the channel attention mechanism module. The stride of the residual module in the upsampling is 1.

[0057] The third upsampling layer and the fourth upsampling layer adopt the same method as the second upsampling layer. After upsampling and concatenation, they are input into the residual module through the channel attention mechanism module. Among them, the input of the third upsampling layer includes the output of the previous upsampling layer and the outputs of the depth convolution path and the standard convolution path of the abnormal brain original image and the texture feature image in the second convolutional layer in the encoder; the input of the fourth upsampling layer includes the output of the previous upsampling layer and the outputs of the depth convolution path and the standard convolution path of the abnormal brain original image and the texture feature image in the first convolutional layer in the encoder. Finally, the softmax function is used to activate the output of the upsampling stage. The channel attention mechanism module and the position attention mechanism module are as Figure 4 and Figure 5 shown.

[0058] Furthermore, the abnormal brain original image and the texture feature image are input into the convolutional neural network. Before input, the collected BraTS2020 brain MR data is divided into patches. The Patch-size is set to 64×64×64, and the stride-patch is set to 32.

[0059] The U-Net model combined with the gray-level co-occurrence matrix and the dual attention mechanism is used to train the training set. After a finite number of iterations, the ideal model is saved. In this embodiment, the model training is set to 20 epochs to complete the training of the model. Optionally, the loss function of the convolutional neural network model is selected as the combination of cross-entropy and dice loss, the optimizer is selected as Adam, and the initial learning rate is selected as le-4.

[0060] When the U-Net network model based on GLCM and the dual attention mechanism is trained, it is tested on the test set and the segmentation effect is evaluated. In this embodiment, the Dice coefficient and the Hausdorff distance are used to evaluate the segmentation result.

[0061] To verify the effectiveness of the improvement of the 3D U-Net network in this embodiment, the abnormal brain segmentation method provided in this application and the medical image segmentation method based on the 3D U-Net network are used to perform model training, verification, and testing on the same training set, validation set, and test set.

[0062] First, the segmentation map is taken out for analysis during the model test. Figure 7Comparison graphs of the segmentation results of two cases of data in the test set segmented by the method provided in this embodiment and the original 3D U-Net model respectively. From Figure 7 As can be seen from Figure 7 , compared with the labels manually segmented by experts, although the detailed information of the edge part is not fine enough, the larger target objects such as tumors, cerebrospinal fluid, white matter, and gray matter can be roughly segmented. Compared with the existing methods for medical image segmentation based on the 3D U-Net network, the abnormal brain segmentation method provided in this embodiment has higher accuracy.

[0063] Secondly, a quantitative analysis is carried out from the Dice similarity coefficient evaluation and Hausdorff distance evaluation indexes of the segmentation results during the model test. Table 1 shows the average Dice results of four segmentation targets, namely tumors, cerebrospinal fluid, white matter, and gray matter, after the test set data is segmented by different convolutional network models; Table 2 shows the average Hausdorff distance results of four segmentation targets, namely tumors, cerebrospinal fluid, white matter, and gray matter, after the test set data is segmented by different convolutional network models. The results are as follows:

[0064] Table 1

[0065]

[0066] Table 2

[0067]

[0068] As can be seen from Table 1 and Table 2, the abnormal brain segmentation method in the embodiment of the present application has a certain improvement in segmentation accuracy, which is also consistent with the above qualitative analysis results. The experimental results show that introducing the gray-level co-occurrence matrix, channel attention mechanism, and position attention mechanism into the U-Net network structure for the segmentation of abnormal brain images can improve the accuracy of the segmentation model.

[0069] Embodiment 2

[0070] Please refer to Figure 8 , Figure 8 which is a schematic structural diagram of the abnormal brain image segmentation system provided in the embodiment of the present application. As shown in Figure 8 , the abnormal brain image segmentation system 800 includes:

[0071] An acquisition module 810, configured to acquire an original abnormal brain image;

[0072] A segmentation module 820 is used to perform normalization processing on the abnormal brain original image, generate a texture feature image using a gray-level co-occurrence matrix, build a convolutional neural network model, input the abnormal brain original image and the texture feature image into the built convolutional neural network model for model training, and segment the nuclear magnetic resonance brain image based on the trained convolutional neural network model.

[0073] Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit; in the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is incorporated in the subsequent upsamplings; the abnormal brain original image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connection.

[0074] Embodiment III

[0075] Please refer to Figure 9 , Figure 9 which is a schematic diagram of a computer device provided by an embodiment of the present application. As Figure 9 shown in, the computer device 900 includes a processor 910, a memory 920, and a bus 930.

[0076] The memory 920 stores machine-readable instructions executable by the processor 910. When the computer device 900 runs, the processor 910 communicates with the memory 920 through the bus 930. When the machine-readable instructions are executed by the processor 910, the steps of the abnormal brain image segmentation method in the method embodiment as shown above can be executed. The specific implementation manner can refer to the method embodiment and will not be elaborated here. Figures 1 to 7 shown,

[0077] Embodiment IV

[0078] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the steps of the abnormal brain image segmentation method in the above method embodiment are executed.

[0079] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0080] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An abnormal brain image segmentation method based on a dual attention mechanism, characterized in that Including: Obtain an abnormal original brain image; Perform normalization processing on the abnormal original brain image, generate a texture feature image using a gray-level co-occurrence matrix, build a convolutional neural network model, calculate the statistical mean, standard deviation, contrast, and dissimilarity using the gray-level co-occurrence matrix, and generate texture feature images of the mean, standard deviation, contrast, and dissimilarity; Input the abnormal original brain image and the texture feature image into the built convolutional neural network model for model training, and segment the magnetic resonance brain image based on the trained convolutional neural network model; Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual module; in the decoder, the first upsampling integrates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is integrated in the subsequent upsampling; The abnormal original brain image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connection; The encoder includes five convolutional modules. The first convolutional module inputs the abnormal original brain image and the generated texture feature image for 3×3×3 convolution, and after passing through the batch normalization layer and the LeakyRelu activation, it is short-connected and input into the next convolutional module. The second to fifth convolutional modules use residual blocks to replace the pooling layer.

2. The abnormal brain image segmentation method according to claim 1, wherein, Also including: Divide the obtained magnetic resonance brain images into a training set and a validation set according to a set ratio; the magnetic resonance brain images in the training set are used to train the convolutional neural network model, and the magnetic resonance brain images in the validation set are used to test and evaluate the segmentation effect of the convolutional neural network model after the convolutional neural network model training is completed.

3. The abnormal brain image segmentation method according to claim 1, characterized in that The residual module includes two units, and each unit consists of regularization, an activation function, and a 3×3×3 convolutional layer.

4. The abnormal brain image segmentation method according to claim 1, wherein The decoder includes four upsampling layers. After the first upsampling and splicing of the feature maps, the spliced channel attention mechanism module and position attention mechanism module are input into the residual module; the input of the second upsampling layer includes the output of the previous upsampling layer and the outputs of the deep convolution path and the standard convolution path of the abnormal original brain image and the texture feature image in the third convolutional layer of the encoder. After the second upsampling and splicing, it is input into the residual module through the channel attention mechanism module; the third and fourth upsampling layers adopt the same method as the second upsampling layer, and finally the softmax function is used to activate the output of the upsampling stage.

5. The abnormal brain image segmentation method according to claim 1, wherein, The loss function of the convolutional neural network model is selected as a combination of cross-entropy and dice loss.

6. An abnormal brain image segmentation system based on a dual attention mechanism, characterized in that, Including: An acquisition module for acquiring an abnormal original brain image; A segmentation module for performing normalization processing on the abnormal original brain image, generating a texture feature image using a gray-level co-occurrence matrix, building a convolutional neural network model, calculating the statistical mean, standard deviation, contrast, and dissimilarity using the gray-level co-occurrence matrix, and generating texture feature images of the mean, standard deviation, contrast, and dissimilarity; Input the abnormal brain original image and the texture feature image into the constructed convolutional neural network model for model training, and segment the magnetic resonance brain image based on the trained convolutional neural network model; Among them, the convolutional neural network model includes an encoder and a decoder. In the encoder, five convolutional modules are set, and each module contains a residual unit; in the decoder, the first upsampling incorporates a position attention mechanism module and a channel attention mechanism module, and only the channel attention mechanism module is incorporated in the subsequent upsamplings; The abnormal brain original image and the texture feature image are respectively subjected to feature extraction by the encoder to obtain feature maps, and the feature maps are input into the decoder after skip connection; The encoder includes five convolutional modules. The first convolutional module inputs the abnormal brain original image and the generated texture feature image for 3×3×3 convolution, and after passing through the batch normalization layer and the LeakyRelu activation, it is short-connected and input into the next convolutional module. The second to fifth convolutional modules use residual blocks to replace the pooling layer.

7. A computer device, characterized in that, Including: A processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the abnormal brain image segmentation method according to any one of claims 1 to 5 are executed.

8. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is run by the processor, the steps of the abnormal brain image segmentation method according to any one of claims 1 to 5 are executed.

Images