Drug user brain magnetic resonance image classification method, system and storage medium

Abstract

The application relates to the technical field of image recognition, in particular to a drug user brain magnetic resonance image classification method and system and a storage medium. A CNN-Mamba hybrid deep learning model with a double-branch parallel structure is used to extract and fuse the features of the brain magnetic resonance image to be classified, and whether the left prefrontal cortex region, the left insular region and the left reward loop related region in the obtained aggregated features meet a preset classification condition is identified; if yes, the brain magnetic resonance image is classified, and reliable classification of the brain magnetic resonance image is realized. The application aims to solve the problem of how to classify drug users according to brain magnetic resonance images.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method, system and storage medium for classifying magnetic resonance images of the brains of drug addicts. Background Technology

[0002] Drug abuse is a growing problem. All types of drugs directly affect the central nervous system, causing irreversible damage to brain nerve cells, and have become one of the major risk factors threatening human health and social stability. Therefore, in-depth research into the underlying reasons why drug addiction is difficult to eradicate and exploring practical intervention methods are not only of significant scientific importance but also contain profound social value. Medical image classification, as an important research direction in the field of medical imaging, plays an indispensable role in clinical practice. With the rapid development of artificial intelligence technology, many researchers are using deep learning models to solve medical image classification problems. Inputting medical images into a deep learning model can complete end-to-end medical image classification and segmentation tasks.

[0003] However, in the field of brain addiction, there are currently some difficulties in using deep learning models to classify brain images of patients addicted to addictive drugs. There is a lack of a method that can accurately capture the brain region features related to addiction in the brain images of such patients.

[0004] In view of this, this application proposes a method for classifying brain magnetic resonance images of drug addicts, aiming to identify and classify drug addicts from brain magnetic resonance images. Summary of the Invention

[0005] The main purpose of this application is to provide a method for classifying brain magnetic resonance images of drug addicts, aiming to solve the problem of how to identify and classify drug addicts from brain magnetic resonance images.

[0006] To achieve the above objectives, this application provides a method for classifying brain magnetic resonance images of drug users, applied to a deep learning model. The deep learning model includes a CNN branch for extracting local features of the brain magnetic resonance images and a Mamba branch for extracting global features of the brain magnetic resonance images. The method includes the following steps: S10, acquire the brain MRI image to be classified; S20, the brain magnetic resonance images to be classified are input into the CNN branch and the Mamba branch respectively to obtain local features and global features respectively and fuse them into a fused feature image; S30, determine the average activation area, left hemisphere lateralization feature and right hemisphere lateralization feature in the fused feature image, wherein the left hemisphere lateralization feature includes the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization feature includes the right prefrontal cortex region and the right insula region. S40, when at least one of the average activation area, the left hemisphere lateralization feature, and the right hemisphere lateralization feature satisfies a preset classification condition, the brain magnetic resonance image to be classified is determined to be a drug user classification image.

[0007] Optionally, the CNN branch adopts a three-layer cascaded structure, wherein: The first layer is a depthwise separable convolution, with the number of input and output channels being half the number of input feature map channels, and the convolution kernel is 3×3; The second layer is a depthwise separable convolution, with the number of input channels being half the number of input feature map channels, and the number of output channels being equal to the number of input feature map channels. The convolution kernel is 3×3. The third layer is a pointwise convolution, with the number of input channels equal to the number of input feature map channels, and the number of output channels being half the number of input feature map channels. The convolution kernel is 1×1. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function.

[0008] Optionally, the Mamba branch includes a parallel visual Mamba block, which includes: The normalization layer is used to normalize the input feature map; The block unit is used to divide the normalized feature map into a first feature map and a second feature map along the channel dimension. A first adaptive visual state space block is used to process the first feature map and output a first intermediate feature. The second adaptive visual state space block is used to process the second feature map and output the second intermediate feature. A splicing unit is used to splice the first intermediate feature and the second intermediate feature along the channel dimension; The channel rearrangement unit is used to rearrange the channels of the concatenated feature map and output globally dependent features.

[0009] Optionally, the adaptive visual state space block processes the input feature map in the following manner: After the input feature map is processed by layer normalization, it is divided into a first branch feature map and a second branch feature map along the channel dimension. In the first branch, the feature map of the first branch is processed sequentially through the first linear layer and the SiLU activation function to generate the output feature of the first branch; In the second branch, the feature map of the second branch is processed sequentially through the second linear layer, the depthwise separable convolutional layer and the SiLU activation function, and then through the two-dimensional scanning module and layer normalization processing to generate the output features of the second branch; The first branch output feature is multiplied element-wise with the second branch output feature to obtain the fused feature; After the fused features are processed by the third linear layer, they are joined with the input feature map through a residual connection that adds element by element, outputting the final features of the adaptive visual state space block.

[0010] Optionally, the two-dimensional scanning module extracts global features by omitting the sampling step and performing intra-group traversal of image blocks.

[0011] Optionally, in step S20, local features and global features are fused into a fused feature image by a feature aggregation module, the feature aggregation module comprising: The first branch is used to downsample and modulate the local features output by the CNN branch to output non-local features; The second branch is used to restore local details of the global dependency features output by the Mamba branch through a multilayer perceptron, and output enhanced local features. The fusion unit is used to concatenate the non-local features and the enhanced local features along the channel dimension, and output the aggregated features after processing by the ECA attention mechanism.

[0012] Optionally, the preset classification conditions include at least one of the following: When the average activation area is less than a preset activation area threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image; When the similarity between the left hemisphere lateralization feature and the preset left hemisphere lateralization reference feature is greater than the preset left hemisphere similarity threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image. When the similarity between the right hemisphere lateralization feature and the preset right hemisphere lateralization reference feature is greater than the preset right hemisphere similarity threshold, the brain MRI image to be classified is determined to be a drug user classification image.

[0013] Furthermore, to achieve the above objectives, this application also provides a brain magnetic resonance imaging (MRI) image classification system, the brain MRI image classification system comprising: The image acquisition module is used to acquire brain magnetic resonance images to be classified. The image processing module is used to input the brain magnetic resonance image to be classified into the CNN branch and the Mamba branch respectively, to obtain local features and global features respectively, and fuse them into a fused feature image; to determine the average activation area, left hemisphere lateralization features and right hemisphere lateralization features in the fused feature image, wherein the left hemisphere lateralization features include the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization features include the right prefrontal cortex region and the right insula region; The image classification module is used to determine that the brain magnetic resonance image to be classified is a drug user classification image when at least one of the average activation area, the left hemisphere lateralization feature and the right hemisphere lateralization feature meets a preset classification condition.

[0014] In addition, to achieve the above objectives, this application also provides a computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for classifying magnetic resonance images of the brains of drug users as described in any of the preceding claims.

[0015] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for classifying magnetic resonance images of the brains of drug users as described in any of the preceding claims.

[0016] This application has at least the following beneficial effects: 1. By constructing a CNN-Mamba hybrid model with a dual-branch parallel structure, effective complementarity between local features and global dependency features is achieved. The CNN branch fully utilizes the efficient local feature extraction capability of depthwise separable convolutions, while the Mamba branch achieves global dependency modeling through adaptive visual state space blocks and an efficient 2D scanning module. These two branches are processed in parallel and fused, achieving excellent classification accuracy in brain magnetic resonance imaging classification tasks.

[0017] 2. Significant progress has been made in model lightweighting. By employing techniques such as depthwise separable convolution, parallel structural design, and efficient 2D scanning modules, the number of model parameters and computational cost of this invention are far lower than those of existing medical image classification models, achieving a balance between lightweighting and high accuracy.

[0018] 3. Use Grad-CAM visualization analysis to find classification indicators and achieve reliable classification of brain MRI images by the model. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the method for classifying magnetic resonance images of the brains of drug users according to embodiments of this application; Figure 2 This is a schematic diagram of the structure of the convolution module involved in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the parallel vision Mamba block involved in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the adaptive vision state space block involved in the embodiments of this application; Figure 5 This is a schematic diagram of the feature aggregation module involved in an embodiment of this application; Figure 6 This is a comparative schematic diagram of Grad-CAM thermal maps involved in the embodiments of this application; The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art. Example 1

[0021] Reference Figure 1 This embodiment provides a method for classifying brain magnetic resonance images of drug users, applied to a deep learning model. The deep learning model includes a CNN branch for extracting local features of the brain magnetic resonance images and a Mamba branch for extracting global features of the brain magnetic resonance images. The method includes the following steps: S10, acquire the brain MRI image to be classified; In this embodiment, the brain MRI image to be classified refers to the image obtained after the brain of the person being tested is acquired by MRI and preprocessed.

[0022] In some alternative implementations, data preprocessing includes the following steps: S11 segments brain MRI images into images of brain tissues such as gray matter, white matter, and cerebrospinal fluid.

[0023] S12 generates a specific template based on the segmentation results and performs image registration to eliminate structural differences between different individuals.

[0024] S13, normalize the registered images to eliminate individual differences.

[0025] S20, the brain magnetic resonance images to be classified are input into the CNN branch and the Mamba branch respectively to obtain local features and global features respectively and fuse them into a fused feature image; In this embodiment, local features of the image are extracted using CNN branches.

[0026] Further and optionally, such as Figure 2 As shown, the CNN branch in this embodiment adopts a three-layer cascaded structure, specifically including: The first layer is a depthwise separable convolution, with both the input and output channels being half the number of channels in the input feature map. The kernel size is 3×3, with padding of 1, and the number of groups equal to the number of input channels. This layer is followed by a batch normalization layer and a ReLU activation function.

[0027] The second layer is a depthwise separable convolution, with the number of input channels being half the number of input feature map channels, and the number of output channels equal to the number of input feature map channels. The convolution kernel size is 3×3, with padding of 1, and the number of groups equal to the number of input channels. This layer is followed by a batch normalization layer and a ReLU activation function.

[0028] The third layer is a pointwise convolution, with the number of input channels equal to the number of input feature map channels, and the number of output channels being half the number of input feature map channels. The convolution kernel size is 1×1. This layer is followed by a batch normalization layer and a ReLU activation function.

[0029] With the above structure, CNN branches can efficiently extract local detail features of an image.

[0030] On the other hand, in this embodiment, global dependency features of the image are extracted through the Mamba branch.

[0031] Further and optionally, such as Figure 3 As shown, the Mamba branch includes a parallel visual Mamba block, which includes: The normalization layer is used to normalize the input feature map; The block unit is used to divide the normalized feature map into a first feature map and a second feature map along the channel dimension. A first adaptive visual state space block is used to process the first feature map and output a first intermediate feature. The second adaptive visual state space block is used to process the second feature map and output the second intermediate feature. A splicing unit is used to splice the first intermediate feature and the second intermediate feature along the channel dimension; The channel rearrangement unit is used to rearrange the channels of the concatenated feature map and output globally dependent features.

[0032] The specific processing procedure is as follows: First, the input feature map is normalized by a layer normalization layer.

[0033] Then, the block unit divides the normalized feature map into a first feature map and a second feature map along the channel dimension.

[0034] Next, the first feature map is input into the first adaptive visual state space block for processing, and the first intermediate feature is output; the second feature map is input into the second adaptive visual state space block for processing, and the second intermediate feature is output. The two adaptive visual state space blocks are set in parallel.

[0035] Then, the splicing unit splices the first intermediate feature and the second intermediate feature along the channel dimension.

[0036] Finally, the channel rearrangement unit rearranges the channels of the concatenated feature map to enhance information exchange between modules and output globally dependent features.

[0037] Further and optionally, such as Figure 4 As shown, this embodiment improves upon the existing VSS module based on the Mamba branch, replacing SS2D with the high-efficiency 2D scanning module ES2D, and proposing an adaptive visual state space block. The specific processing procedure is as follows: After layer normalization, the input feature map M_apx is segmented along the channel dimension into the first branch feature map M. apk1 Second branch feature map M apk2 : ; In the first branch, the first branch feature map M apk1 The first branch output feature M' is generated by sequentially processing the first linear layer and the SiLU activation function. apk1 : ; In the second branch, the feature map M of the second branch apk2 After being processed sequentially through a second linear layer, a depthwise separable convolutional layer, and a SiLU activation function, and then through an efficient 2D scanning module and layer normalization, the second branch output feature M'' is generated. apk1 : ; The output features of the first branch and the output features of the second branch are multiplied element-wise to obtain the fused feature M. apk : ;

[0038] After the fused features are processed by the third linear layer, they are joined with the input feature map through a residual connection that adds element-wise, outputting the final feature M of the adaptive visual state space block. apG: ; Where Linear(·) represents linear layer processing, ES2D(·) represents an efficient 2D scanning module, DWConv(·) represents depthwise separable convolution, and SiLU(·) represents the SiLU activation function. ⊕ indicates element-wise multiplication, and ⊕ indicates element-wise addition.

[0039] It should be noted that the high-efficiency 2D scanning module extracts global features by omitting the sampling step and performing intra-group traversal of image blocks, thereby reducing the amount of computation while maintaining the integrity of global information.

[0040] Further and optionally, in this step, local and global features are fused into a fused feature image using the proposed feature aggregation module. For example... Figure 5 As shown, the feature aggregation module adopts a dual-branch structure: The first branch is used to downsample and modulate the local features output by the CNN branch, and output non-local features X1.

[0041] The second branch is used to restore local details of the global dependency features output by the Mamba branch through DMlp, and outputs the enhanced local feature Yd.

[0042] The fusion unit concatenates non-local features and enhanced local features along the channel dimension, and outputs aggregated features after processing by the ECA attention mechanism.

[0043] Specifically, the process can be represented as:

[0044] Where FL represents local features and FG represents global dependency features. This indicates splicing along the channel dimension.

[0045] In this step, after four stages of feature extraction, the deep learning model outputs the classification result through a classification head. Optionally, the classification head includes, in sequence: A 1×1 convolutional layer is used to adjust the number of channels in the feature map, mapping the feature map to the channel dimension required for classification.

[0046] The global pooling layer is used to compress the spatial dimension to 1×1 and obtain a global feature representation.

[0047] The fully connected layer is used to map the globally pooled feature vectors to the number of categories, and output the final classification result.

[0048] S30, determine the average activation area, left hemisphere lateralization feature and right hemisphere lateralization feature in the fused feature image, wherein the left hemisphere lateralization feature includes the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization feature includes the right prefrontal cortex region and the right insula region. In this embodiment, to clarify the changes in brain location among drug users, we collaborated with a judicial drug rehabilitation center in Yunnan Province to collect brain MRI images between 2023 and 2024, collecting data from 177 drug users and 110 healthy controls. All subjects were between 18 and 60 years old. Data acquisition was performed using a United Imaging 3.0T MRI scanner with T1-weighted imaging sequences. The data acquisition process followed the ethical principles of the Declaration of Helsinki, and all subjects signed informed consent forms. Brain MRI image data of the patients were collected. Under the guidance of professional physicians, image slices from layers 20 to 104 were selected from 120 layers of images for each subject as experimental data. The Grad-CAM algorithm was used to visualize and analyze the model's classification results, generating heatmaps to display the brain regions of interest to the model, as shown below. Figure 6 A comparative analysis of the heatmaps of healthy samples and drug-addicted samples was conducted. The following conclusions were drawn: First, the heatmaps of healthy samples exhibit a bilaterally symmetrical activation pattern, with highlighted areas (red / yellow) symmetrically distributed in both hemispheres or concentrated in the midline brain region. Quantitative analysis shows that the average activation area ratio of healthy samples is 35.16%, while the average activation area ratio of drug-using samples is 24.99%, a decrease of 28.9% compared to healthy samples. Based on this, we selected the average activation area as one of the classification indicators, using whether the average activation area in the acquired images is less than a preset threshold to determine whether the image corresponds to the target classification brain image of the person being tested.

[0049] Second, the heatmaps of drug-using samples showed a significant left-hemispheric lateralization, with highlighted areas mainly concentrated in the left prefrontal cortex, left insula, and brain regions associated with the left-side reward circuit. The underlying principle is that exposure to chronic addictive drugs easily leads to sensitization of the left-hemispheric reward circuit, while simultaneously causing a moderate reduction in the activation range of the overall left prefrontal cortex and left insula.

[0050] Based on this, we selected the left prefrontal cortex region, the left insula region, and the left reward circuit-related area as one of the classification indicators, enabling the deep learning model to learn the features of these regions in the brain image of drug addicts for classification.

[0051] Third, a small number of drug user samples showed a bright spot in the right hemisphere. The underlying principle is that the bright spot in the right prefrontal cortex, as the core brain region for inhibitory control, reflects the compensatory effect after the executive function is impaired; the bright spot in the right insula is related to the immediate craving state at the time of sample collection.

[0052] Based on this, we selected the right prefrontal cortex and the right insula region as one of the classification indicators, enabling the deep learning model to learn the features of these regions in the brain image of drug users for classification.

[0053] S40, when at least one of the average activation area, the left hemisphere lateralization feature, and the right hemisphere lateralization feature satisfies a preset classification condition, the brain magnetic resonance image to be classified is determined to be a drug user classification image.

[0054] In this step, further and optionally, satisfying the preset classification conditions includes at least one of the following: 1. When the average activation area is less than a preset activation area threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image; 2. When the similarity between the left hemisphere lateralization feature and the preset left hemisphere lateralization reference feature is greater than the preset left hemisphere similarity threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image. Third, when the similarity between the right hemisphere lateralization feature and the preset right hemisphere lateralization reference feature is greater than the preset right hemisphere similarity threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image.

[0055] It should be noted that the classification method involved in this embodiment only involves the target classification of brain magnetic resonance images, and does not involve the diagnosis of the test subject based on brain magnetic resonance images.

[0056] In the technical solution provided in this embodiment, a CNN-Mamba hybrid deep learning model with a dual-branch parallel structure is used to extract and fuse features from brain magnetic resonance images to be classified. The model identifies whether the left prefrontal cortex region, left insula region, and left reward circuit related region in the obtained aggregated features meet the preset classification adjustment. If they meet the requirements, they are classified to achieve reliable classification of brain magnetic resonance images.

[0057] Verification Implementation Examples Based on the first embodiment, this embodiment uses a five-fold cross-validation method to evaluate the model performance. Specifically, the dataset is divided into a training set and a validation set using multiple random seeds, with the training set accounting for 80% and the validation set accounting for 20%. Five random partitioning experiments are conducted, and the average result of the five experiments is used as the performance evaluation criterion.

[0058] The experiment used accuracy, precision, recall, and F1 score as evaluation metrics. The proposed CNN-Mamba hybrid model was compared with existing classic CNN models (including ResNet34, MobileNetV2, ShuffleNet, DenseNet, EfficientNet, RegNet, and EfficientNetV2) and existing Mamba models (including MedMamba and MicMamba).

[0059] The experimental results are shown in Table 1: ; It can be seen that: First, among classic CNN models, EfficientNetV2 achieved the best overall performance, with a recall rate of 0.880. Compared to other CNN models, EfficientNetV2 is more suitable for medical scenarios in clinical diagnosis, but its performance still has room for improvement.

[0060] Second, in the Mamba model, MedMamba and MicMamba each have their advantages. Both are suitable for medical scenarios, but MicMamba performs better in local feature processing, indicating that the combination of CNN and Mamba is feasible.

[0061] Third, the CNN-Mamba model proposed in this application outperforms the classic CNN model and the Mamba model on all evaluation metrics, achieving an accuracy of 0.907, a precision of 0.881, a recall of 0.898, and an F1-score of 0.889. Regarding model complexity, the model of this invention has 2.8M parameters and a computational cost of 0.64G FLOPs, which are 78.8% and 68.2% lower than the existing medical image model MedMamba (13.3M parameters, 2.04G FLOPs), respectively, achieving a balance between lightweight design and high accuracy.

[0062] Furthermore, as one implementation, this application embodiment also relates to a brain magnetic resonance imaging (MRI) image classification system, the brain MRI image classification system comprising: The image acquisition module is used to acquire brain magnetic resonance images to be classified. The image processing module is used to input the brain magnetic resonance image to be classified into the CNN branch and the Mamba branch respectively, to obtain local features and global features respectively, and fuse them into a fused feature image; to determine the average activation area, left hemisphere lateralization features and right hemisphere lateralization features in the fused feature image, wherein the left hemisphere lateralization features include the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization features include the right prefrontal cortex region and the right insula region; The image classification module is used to determine that the brain magnetic resonance image to be classified is a drug user classification image when at least one of the average activation area, the left hemisphere lateralization feature and the right hemisphere lateralization feature meets a preset classification condition.

[0063] Furthermore, as an implementation scheme, the present application embodiment also relates to a computer system, the computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for classifying magnetic resonance images of the brain of drug addicts as described above.

[0064] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.

[0065] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the drug user brain magnetic resonance image classification method as described in the above embodiments.

[0066] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0067] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0068] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0073] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for classifying magnetic resonance imaging (MRI) images of the brains of drug addicts, characterized in that, The method, applied to a deep learning model including a CNN branch for extracting local features from brain MRI images and a Mamba branch for extracting global features from the brain MRI images, includes the following steps: S10, acquire the brain MRI image to be classified; S20, the brain magnetic resonance images to be classified are input into the CNN branch and the Mamba branch respectively to obtain local features and global features respectively and fuse them into a fused feature image; S30, determine the average activation area, left hemisphere lateralization feature and right hemisphere lateralization feature in the fused feature image, wherein the left hemisphere lateralization feature includes the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization feature includes the right prefrontal cortex region and the right insula region. S40, when at least one of the average activation area, the left hemisphere lateralization feature, and the right hemisphere lateralization feature satisfies a preset classification condition, the brain magnetic resonance image to be classified is determined to be a drug user classification image.

2. The method for classifying magnetic resonance images of the brains of drug users as described in claim 1, characterized in that, The CNN branch adopts a three-layer cascaded structure, wherein: The first layer is a depthwise separable convolution, with the number of input and output channels being half the number of input feature map channels, and the convolution kernel is 3×3; The second layer is a depthwise separable convolution, with the number of input channels being half the number of input feature map channels, and the number of output channels being equal to the number of input feature map channels. The convolution kernel is 3×3. The third layer is a pointwise convolution, with the number of input channels equal to the number of input feature map channels, and the number of output channels being half the number of input feature map channels. The convolution kernel is 1×1. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function.

3. The method for classifying magnetic resonance images of the brains of drug users as described in claim 1, characterized in that, The Mamba branch includes a parallel visual Mamba block, which includes: The normalization layer is used to normalize the input feature map; The block unit is used to divide the normalized feature map into a first feature map and a second feature map along the channel dimension. A first adaptive visual state space block is used to process the first feature map and output a first intermediate feature. The second adaptive visual state space block is used to process the second feature map and output the second intermediate feature. A splicing unit is used to splice the first intermediate feature and the second intermediate feature along the channel dimension; The channel rearrangement unit is used to rearrange the channels of the concatenated feature map and output globally dependent features.

4. The method for classifying magnetic resonance images of the brains of drug users as described in claim 3, characterized in that, The adaptive visual state space block processes the input feature map in the following way: After the input feature map is processed by layer normalization, it is divided into a first branch feature map and a second branch feature map along the channel dimension. In the first branch, the feature map of the first branch is processed sequentially through the first linear layer and the SiLU activation function to generate the output feature of the first branch; In the second branch, the feature map of the second branch is processed sequentially through the second linear layer, the depthwise separable convolutional layer and the SiLU activation function, and then through the two-dimensional scanning module and layer normalization processing to generate the output feature of the second branch; The first branch output feature is multiplied element-wise with the second branch output feature to obtain the fused feature; After the fused features are processed by the third linear layer, they are connected with the input feature map through a residual connection that adds element by element, and the final features of the adaptive visual state space block are output.

5. The method for classifying magnetic resonance images of the brains of drug users as described in claim 4, characterized in that, The two-dimensional scanning module extracts global features by omitting the sampling step and performing intra-group traversal of image blocks.

6. The method for classifying magnetic resonance images of the brains of drug users as described in claim 1, characterized in that, In step S20, local and global features are fused into a fused feature image through a feature aggregation module. The feature aggregation module includes: The first branch is used to downsample and modulate the local features output by the CNN branch to output non-local features; The second branch is used to restore local details of the global dependency features output by the Mamba branch through a multilayer perceptron, and output enhanced local features. The fusion unit is used to concatenate the non-local features and the enhanced local features along the channel dimension, and output the aggregated features after processing by the ECA attention mechanism.

7. The method for classifying magnetic resonance images of the brains of drug users as described in claim 1, characterized in that, The preset classification conditions include at least one of the following: When the average activation area is less than a preset activation area threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image; When the similarity between the left hemisphere lateralization feature and the preset left hemisphere lateralization reference feature is greater than the preset left hemisphere similarity threshold, the brain magnetic resonance image to be classified is determined to be a drug user classification image. When the similarity between the right hemisphere lateralization feature and the preset right hemisphere lateralization reference feature is greater than the preset right hemisphere similarity threshold, the brain MRI image to be classified is determined to be a drug user classification image.

8. A brain magnetic resonance imaging classification system, characterized in that, The brain magnetic resonance imaging classification system includes: The image acquisition module is used to acquire brain magnetic resonance images to be classified. The image processing module is used to input the brain magnetic resonance image to be classified into the CNN branch and the Mamba branch respectively, to obtain local features and global features respectively, and fuse them into a fused feature image; to determine the average activation area, left hemisphere lateralization features and right hemisphere lateralization features in the fused feature image, wherein the left hemisphere lateralization features include the left prefrontal cortex region, the left insula region and the left reward circuit related area, and the right hemisphere lateralization features include the right prefrontal cortex region and the right insula region; The image classification module is used to determine that the brain magnetic resonance image to be classified is a drug user classification image when at least one of the average activation area, the left hemisphere lateralization feature and the right hemisphere lateralization feature meets a preset classification condition.

9. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for classifying magnetic resonance images of the brains of drug users as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for classifying magnetic resonance images of the brains of drug users as described in any one of claims 1 to 7.

Images