MRI image classification method, system and terminal based on brain age information guidance
By constructing a brain age information-guided MRI image classification method, using the feature fusion of brain age prediction branches and AD prediction branches, combined with the brain age guidance attention module, the problem of failure to effectively utilize brain age information in the existing technology is solved, and high-precision classification of MRI images is achieved.
Patent Information
- Application Number
- CN202510208635.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-07-25
AI Technical Summary
The existing MRI image classification methods fail to effectively utilize brain age information, resulting in insufficient classification accuracy of deep network models.
A MRI image classification method based on brain age information guidance is constructed, including brain age prediction branches, AD prediction branches, brain age guidance attention module and full connection layer. Through preprocessing, feature extraction and feature fusion, brain age knowledge can be used to improve classification effect.
It effectively improves the classification accuracy of MRI images, makes full use of the complementary advantages of brain age information and network architecture, and achieves robust image classification.
Smart Images
Figure CN120375031A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing, and in particular, to an MRI image classification method, system, terminal, and computer-readable storage medium guided by brain age information. Background Art
[0002] Brain age is an important biomarker reflecting the structural and functional status of the brain. It evaluates the "biological age" of an individual's brain based on brain imaging data through computer algorithms. Compared with an individual's physiological age, brain age can more accurately reflect the degree of brain aging and health status. The difference between the predicted brain age and the actual age, that is, the brain age difference, has been proven to be an early warning indicator for various nervous systems. Research shows that when the predicted brain age is significantly greater than the actual age, it often indicates that the brain may have accelerated aging or structural changes.
[0003] MRI (Magnetic Resonance Imaging) is a non-invasive neuroimaging examination method that can detect fine structural changes in brain regions and contains effective information related to brain age, which can be used to achieve brain age prediction. However, in existing MRI image classification methods, little use is made of the brain age information contained in MRI images. Even if brain age information is used, it cannot be effectively applied to the classification of MRI images, resulting in insufficient classification accuracy of the current deep network model for MRI images.
[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention
[0005] The main objective of the present invention is to provide an MRI image classification method, system, terminal, and computer-readable storage medium guided by brain age information, aiming to solve the problem in the existing technology that the brain age information contained in MRI images cannot be well utilized to achieve MRI image classification, resulting in insufficient classification accuracy of the current deep network model for MRI images.
[0006] To achieve the above objective, the present invention provides an MRI image classification method guided by brain age information. The MRI image classification method guided by brain age information includes the following steps:
[0007] Obtain the MRI image and actual age of a target object, preprocess the MRI image to obtain a target MRI image;
[0008] Construct an MRI image classification model, train and test the MRI image classification model to obtain a target model. The target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer;
[0009] Input the target MRI image into the brain age prediction branch of the target model for prediction to output a predicted brain age value, and input the target MRI image into the AD prediction branch of the target model for prediction to output AD-related features;
[0010] Input the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain a final feature map, and pass the final feature map through the fully connected layer to obtain the classification result of the target MRI image.
[0011] Optionally, in the MRI image classification method guided by brain age information, the brain age prediction branch includes a first PEB module and four-stage CRB modules;
[0012] The step of inputting the target MRI image into the brain age prediction branch of the target model for prediction to output a predicted brain age value specifically includes:
[0013] Input the target MRI image into the first PEB module for segmentation and encoding to obtain brain age encoded features;
[0014] Input the brain age encoded features into the CRB module in the first stage of the brain age prediction branch for Token fusion and channel fusion to obtain fused features, where the fused features include a first fused feature and a second fused feature;
[0015] Input the first fused feature and the second fused feature into the CRB module in the next stage to continue Token fusion and channel fusion until, after fusion by all CRB modules, a predicted brain age value is obtained.
[0016] Optionally, in the MRI image classification method guided by brain age information, the CRB module includes a first CRB architecture and a second CRB architecture;
[0017] The step of inputting the target MRI image into the CRB module in the first stage of the brain age prediction branch for Token fusion and channel fusion to obtain fused features specifically includes:
[0018] When the stride is 1, input the target MRI image into the first CRB architecture, perform downsampling on the target MRI image through parallel depth convolutional layers and residual layers, input the result of downsampling into a batch normalization layer for normalization processing, and integrate the normalized result through CCA to obtain a first Token fused feature;
[0019] Output the first Token fusion feature to an inverted bottleneck structure for channel feature extraction and fusion to obtain a first fusion feature, where the inverted bottleneck structure is, from left to right: a pointwise convolution layer, a first residual layer, a GeLU activation function, a pointwise convolution layer, and a second residual layer;
[0020] When the stride is 2, input the target MRI image into the second CRB architecture, perform downsampling on the target MRI image through a depth convolution layer, input the result of the downsampling into a batch normalization layer for normalization processing, integrate the result of the normalization through CCA, and double the dimension of the integrated result through a pointwise convolution layer to obtain a second Token fusion feature;
[0021] Output the second Token fusion feature to an inverted bottleneck structure for channel feature extraction and fusion to obtain a second fusion feature.
[0022] Optionally, in the MRI image classification method guided by brain age information, the processing process of the CCA is as follows:
[0023] Input the input feature map into three different channels for average pooling respectively to obtain a first coordinate intermediate feature map, a second coordinate intermediate feature map, and a third coordinate intermediate feature map, rearrange the first coordinate intermediate feature map and the third coordinate intermediate feature map, and splice the rearranged first coordinate intermediate feature map, the third coordinate intermediate feature map, and the second coordinate intermediate feature map to obtain a first coordinate splicing result;
[0024] Input the first coordinate splicing result into a CBS module for processing to obtain a second coordinate splicing result, split the second coordinate splicing result to obtain three split feature maps, and pass the three split feature maps through a point convolution layer and a sigmoid activation function to obtain a first coordinate feature, a second coordinate feature, and a third coordinate feature;
[0025] Input the input feature map into three different channels for average pooling, max pooling, and range pooling respectively to obtain a first channel intermediate feature map, a second channel intermediate feature map, and a third channel intermediate feature map, and splice the first channel intermediate feature map, the second channel intermediate feature map, and the third channel intermediate feature map to obtain a first channel splicing result;
[0026] Input the first channel splicing result into a CBS module for processing to obtain a second channel splicing result, and split the second coordinate splicing result to obtain a fourth channel intermediate feature map, a fifth channel intermediate feature map, and a sixth channel intermediate feature map;
[0027] Convolve the intermediate feature map of the fourth channel and the intermediate feature map of the fifth channel through a point convolution layer, add the elements of the convolved intermediate feature map of the fourth channel and the convolved intermediate feature map of the fifth channel to obtain the intermediate feature map of the seventh channel. Convolve the intermediate feature map of the sixth channel through a point convolution layer, multiply the elements of the convolved intermediate feature map of the sixth channel and the intermediate feature map of the seventh channel to obtain the intermediate feature map of the eighth channel. Pass the intermediate feature map of the eighth channel through the sigmoid activation function to obtain the channel feature map;
[0028] Perform an element-wise outer product of the first coordinate feature, the second coordinate feature, the third coordinate feature, the channel feature map, and the input feature map to obtain the output feature map.
[0029] Optionally, in the MRI image classification method guided by brain age information, the CBS module includes a convolution layer, a normalization layer, and a swish activation function.
[0030] Optionally, in the MRI image classification method guided by brain age information, the AD prediction branch includes a second PEB module, two-stage CCB modules, and two-stage DB modules;
[0031] Input the target MRI image into the AD prediction branch of the target model for prediction to output AD-related features, specifically including:
[0032] Input the target MRI image into the second PEB module for segmentation and encoding to obtain AD-encoded features;
[0033] Input the AD-encoded features into the CCB module for calculation to obtain CCB output features, and synchronously obtain the fusion features from the CRB module. Multiply the fusion features by the corresponding learnable coefficients to obtain the first learned features, and add the first learned features and the CCB output features to obtain AD intermediate features;
[0034] Input the AD intermediate features into the DB module for calculation to obtain DB output features, and synchronously obtain the fusion features from the CRB module. Multiply the fusion features by the corresponding learnable coefficients to obtain the second learned features, and add the second learned features and the DB output features to obtain AD-related features.
[0035] Optionally, in the MRI image classification method guided by brain age information, the integration of the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model to obtain the final feature map specifically includes:
[0036] Calculate the brain age difference based on the actual age and the predicted brain age value, and generate the brain age attention weight through the fully connected layer and the sigmoid function of the brain age guided attention module;
[0037] Input the AD-related features into the brain age guided attention module for average pooling, and multiply the result of the average pooling by the brain age attention weight to obtain the final feature map.
[0038] In addition, to achieve the above object, the present invention also provides an MRI image classification system guided by brain age information, wherein the MRI image classification system guided by brain age information includes:
[0039] A target image acquisition module, configured to acquire the MRI image and the actual age of a target object, preprocess the MRI image to obtain a target MRI image;
[0040] A target model construction module, configured to construct an MRI image classification model, train and test the MRI image classification model to obtain a target model, and the target model includes: a brain age prediction branch, an AD prediction branch, a brain age guided attention module, and a fully connected layer;
[0041] A dual-branch feature extraction module, configured to input the target MRI image into the brain age prediction branch of the target model for prediction, output a predicted brain age value, input the target MRI image into the AD prediction branch of the target model for prediction, and output AD-related features;
[0042] A feature fusion classification module, configured to input the actual age, the predicted brain age value, and the AD-related features into the brain age guided attention module of the target model for integration to obtain a final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
[0043] In addition, to achieve the above object, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a brain age information-guided MRI image classification program stored on the memory and executable on the processor, and when the brain age information-guided MRI image classification program is executed by the processor, the steps of the above-mentioned brain age information-guided MRI image classification method are implemented.
[0044] In addition, to achieve the above object, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a brain age information-guided MRI image classification program, and when the brain age information-guided MRI image classification program is executed by a processor, the steps of the above-mentioned brain age information-guided MRI image classification method are implemented.
[0045] In the present invention, an MRI image and the actual age of a target object are obtained, the MRI image is preprocessed to obtain a target MRI image; an MRI image classification model is constructed, the MRI image classification model is trained and tested to obtain a target model, and the target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer; the target MRI image is input into the brain age prediction branch for prediction, and a predicted brain age value is output, the target MRI image is input into the AD prediction branch for prediction, and AD-related features are output; the actual age, the predicted brain age value, and the AD-related features are input into the brain age-guided attention module for integration to obtain a final feature map, and the classification result of the target MRI image is obtained after passing the final feature map through the fully connected layer. The present invention utilizes the complementary advantages of brain age knowledge and network architecture, and effectively improves the classification effect of MRI images. Description of the Drawings
[0046] Figure 1 is a flowchart of a preferred embodiment of the MRI image classification method guided by brain age information according to the present invention;
[0047] Figure 2 is an overall architecture diagram of the target model in the MRI image classification method guided by brain age information according to the present invention;
[0048] Figure 3 is a structural diagram of the CRB module in the MRI image classification method guided by brain age information according to the present invention;
[0049] Figure 4 is a schematic diagram of the processing process of CCA in the MRI image classification method guided by brain age information according to the present invention;
[0050] Figure 5 is a structural diagram of the CCB module in the MRI image classification method guided by brain age information according to the present invention;
[0051] Figure 6 is a structural diagram of the DB module in the MRI image classification method guided by brain age information according to the present invention;
[0052] Figure 7 is a structural diagram of a preferred embodiment of the MRI image classification system guided by brain age information according to the present invention;
[0053] Figure 8 is a structural diagram of a preferred embodiment of the terminal according to the present invention. Detailed Embodiments
[0054] The present application provides an MRI image classification method, system and terminal guided by brain age information. To make the objectives, technical solutions and effects of the present application clearer and more definite, the following further describes the present application in detail with reference to the accompanying drawings and by way of examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0055] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the field to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.
[0056] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the technical features indicated. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions appears to be contradictory or unable to be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.
[0057] The MRI image classification method guided by brain age information according to a preferred embodiment of the present invention is as Figure 1 shown. The MRI image classification method guided by brain age information includes the following steps:
[0058] Step S10: Obtain the MRI image and actual age of the target object, and preprocess the MRI image to obtain a target MRI image.
[0059] Specifically, obtain the MRI image and actual age of the target object to be tested. Among them, the MRI image is used to input the network for feature extraction, and the actual age is used to calculate the brain age difference. Further, preprocess the MRI image. The preprocessing process includes preprocessing techniques such as random cropping, data cleaning, and data standardization, and finally obtain a target MRI image for inputting into the model.
[0060] Step S20: Construct an MRI image classification model, train and test the MRI image classification model to obtain a target model. The target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer.
[0061] It is understandable that before training and testing the MRI image classification model, data needs to be collected to construct a dataset for training the model. In this embodiment, historical data is obtained, and the historical data includes historical MRI images and the classification results corresponding to the historical MRI images; the image size of the historical MRI images is adjusted to pixels of a preset size (for example, 113×137×113) to keep the image sizes consistent, and Statistical Parametric Mapping (SPM) is used to preprocess the MRI images after size adjustment. Finally, the MRI images after correction and skull stripping are used as the training samples input to the network, and the classification results corresponding to the historical retinal images are used as labels to construct a dataset; the dataset is divided into a training set, a test set, and a validation set according to a preset ratio. The training set is used to train the MRI image classification model, the test set is used to evaluate the MRI image classification model in each round of training, and the validation set is used to evaluate the trained MRI image classification model.
[0062] The MRI image classification model is trained and tested to obtain a target model. Specifically, the brain age prediction branch is pre-trained using the dataset. The training set, validation set, and test set are divided in a ratio of 8:1:1, the learning rate is 0.001, the cosine annealing learning rate update strategy is used, AdamW (a variant of the Adam optimizer) is used as the optimizer, and the Mean Absolute Error (MAE) is used as the loss function. A total of 30 rounds of training are performed, and the model weights with the minimum MAE are saved. Then, the weights with the minimum MAE are reloaded into the brain age prediction branch and transferred learning is performed on the samples. The hyperparameters used in this process are the same as those in the pre-training, and the weights with the minimum MAE are also saved. Finally, these weights are loaded into the brain age prediction branch and the weight parameters are frozen to give information related to the brain age, including the predicted brain age and the related multi-scale feature maps. All the remaining data samples are used for the subsequent binary classification training, and five-fold cross-validation is used to verify the performance of the model. The input images will enter the brain age prediction branch and the AD prediction branch simultaneously. The AD prediction branch will fuse the features from the brain age prediction branch and give the final AD classification result at the fully connected layer.
[0063] Further, as Figure 2 shown, the target model consists of a brain age prediction branch with frozen weights, an AD prediction branch with trainable weights, a brain age-guided attention module, and a fully connected layer for classification. The images input to the network are input into two parallel branches.
[0064] The brain age prediction branch consists of a CRB (CCA-RepViT) with frozen weights, which processes the input image to generate a predicted brain age and related multi-scale feature maps. The AD (Alzheimer's disease) prediction branch uses a CCA-CDNet with trainable weights to fuse the feature maps and generate AD-related features. Finally, the BrainAge Guided Attention Module (BAGAM) further integrates the outputs of the two branches and converts the brain age information into brain age attention to guide AD classification. This dual-branch method effectively utilizes brain age knowledge to improve the accuracy of AD classification.
[0065] Step S30: Input the target MRI image into the brain age prediction branch of the target model for prediction, output the predicted brain age value, and input the target MRI image into the AD prediction branch of the target model for prediction, output AD-related features.
[0066] It can be understood that the brain age prediction branch includes a first PEB (3D Patch Embedding) module and four stages of CRB modules;
[0067] The step of inputting the target MRI image into the brain age prediction branch of the target model for prediction and outputting the predicted brain age value specifically includes:
[0068] Input the target MRI image into the first PEB module for segmentation and encoding to obtain brain age encoded features;
[0069] Input the brain age encoded features into the CRB module of the first stage in the brain age prediction branch for Token fusion and channel fusion to obtain fused features, where the fused features include a first fused feature and a second fused feature;
[0070] Input the first fused feature and the second fused feature into the CRB module of the next stage to continue Token fusion and channel fusion until after all CRB module fusions, obtain the predicted brain age value.
[0071] It can be understood that the CRB module includes a first CRB architecture and a second CRB architecture.
[0072] The step of inputting the target MRI image into the CRB module of the first stage in the brain age prediction branch for Token fusion and channel fusion to obtain fused features specifically includes:
[0073] When the step size is 1, the target MRI image is input into the first CRB architecture. The target MRI image is downsampled through parallel depth convolution layers and residual layers, and the result of downsampling is input into a batch normalization layer for normalization processing. The normalized result is integrated through CCA to obtain the first Token fusion feature;
[0074] The first Token fusion feature is output to an inverted bottleneck structure for channel feature extraction and fusion to obtain the first fusion feature. Among them, the inverted bottleneck structure from left to right is: a pointwise convolution layer, a first residual layer, a GeLU activation function, a pointwise convolution layer, and a second residual layer;
[0075] When the step size is 2, the target MRI image is input into the second CRB architecture. The target MRI image is downsampled through a depth convolution layer, and the result of downsampling is input into a batch normalization layer for normalization processing. The normalized result is integrated through CCA, and the integrated result is doubled in dimension through a pointwise convolution layer to obtain the second Token fusion feature;
[0076] The second Token fusion feature is output to an inverted bottleneck structure for channel feature extraction and fusion to obtain the second fusion feature.
[0077] In this embodiment, the brain age prediction branch is constructed by a four-stage CRB with frozen weights, which is used to predict the brain age and generate a multi-scale feature map that can be fused into the AD branch. As Figure 3 shown, Figure 3 shows the architecture of the CRB. It includes two main steps: Token fusion and channel fusion. The behavior of Token fusion depends on the step size. When the step size is 2, depth convolution (DW) is used for downsampling, and the channel dimension is doubled through PW. When the step size is 1, parallel DW and residual modules are adopted. Subsequently, CCA is used to integrate the features between different channels and coordinates. Channel fusion adopts an inverted bottleneck structure, which can extract more complex and information-rich channel features.
[0078] Furthermore, the processing process of the CCA is as follows:
[0079] The input feature map is respectively input into three different channels for average pooling to obtain the first coordinate intermediate feature map, the second coordinate intermediate feature map, and the third coordinate intermediate feature map. The first coordinate intermediate feature map and the third coordinate intermediate feature map are rearranged, and the rearranged first coordinate intermediate feature map, the third coordinate intermediate feature map, and the second coordinate intermediate feature map are concatenated to obtain the first coordinate concatenation result;
[0080] Input the first coordinate splicing result into the CBS module for processing to obtain the second coordinate splicing result. Split the second coordinate splicing result to obtain three split feature maps. Pass the three split feature maps through a point convolution layer and a sigmoid activation function to obtain the first coordinate feature, the second coordinate feature, and the third coordinate feature;
[0081] Input the input feature map into three different channels respectively for average pooling, max pooling, and range pooling to obtain the first-channel intermediate feature map, the second-channel intermediate feature map, and the third-channel intermediate feature map. Concatenate the first-channel intermediate feature map, the second-channel intermediate feature map, and the third-channel intermediate feature map to obtain the first-channel concatenation result;
[0082] Input the first-channel concatenation result into the CBS module for processing to obtain the second-channel concatenation result. Split the second coordinate concatenation result to obtain the fourth-channel intermediate feature map, the fifth-channel intermediate feature map, and the sixth-channel intermediate feature map;
[0083] Convolve the fourth-channel intermediate feature map and the fifth-channel intermediate feature map through a point convolution layer, add the elements of the convolved fourth-channel intermediate feature map and the convolved fifth-channel intermediate feature map to obtain the seventh-channel intermediate feature map. Convolve the sixth-channel intermediate feature map through a point convolution layer, multiply the elements of the convolved sixth-channel intermediate feature map and the seventh-channel intermediate feature map to obtain the eighth-channel intermediate feature map. Pass the eighth-channel intermediate feature map through a sigmoid activation function to obtain the channel feature map;
[0084] Perform an element-wise outer product of the first coordinate feature, the second coordinate feature, the third coordinate feature, the channel feature map, and the input feature map to obtain the output feature map.
[0085] Among them, the CBS module in CCA includes a convolution layer, a normalization layer, and a swish activation function. CBS (convolution-convolution-swish, convolution-normalization-activation function module).
[0086] As Figure 4 shown, Figure 4 illustrates the working mechanism of CCA (Channel-Coordinate Attention, channel-coordinate attention mechanism). Inspired by coordinate attention, it is extended to a 3D form in this implementation to achieve calculations in the depth, width, and height dimensions. In addition, inspired by the convolutional block attention module, range pooling along the channel dimension is further introduced, which returns the difference between the maximum and minimum values within each channel, reflecting the richness of the feature map. CCA generates voxel-level attention maps for both channel and coordinate features, improving the computational efficiency.
[0087] Further, the AD prediction branch includes a second PEB module, CCB modules in two stages, and DB modules in two stages;
[0088] Input the target MRI image into the AD prediction branch of the target model for prediction, and output AD-related features, specifically including:
[0089] Input the target MRI image into the second PEB module for segmentation and encoding to obtain AD encoded features;
[0090] Input the AD encoded features into the CCB module for calculation to obtain CCB output features, and simultaneously obtain the fused features from the CRB module. Multiply the fused features by the corresponding learnable coefficients to obtain the first learned features, and add the first learned features to the CCB output features to obtain AD intermediate features;
[0091] Input the AD intermediate features into the DB module for calculation to obtain DB output features, and simultaneously obtain the fused features from the CRB module. Multiply the fused features by the corresponding learnable coefficients to obtain the second learned features, and add the second learned features to the DB output features to obtain AD-related features.
[0092] In this embodiment, the AD prediction branch uses CCA-CDNet with trainable weights as the backbone network. CCA-CDNet multiplies the multi-scale feature maps by the learnable coefficient α i and then sums them to generate AD-related features. CCA-CDNet is constructed by CCB (3DCCA-ConvNeXtV2) and DB (3D DaViT). In this embodiment, two CCB stages are adopted, followed by two DB stages (as Figure 2 shown).
[0093] As Figure 5 shown, Figure 5 shows the structure diagram of the CCB, including a downsampling layer, a depth convolution layer, a normalization layer, a CCA layer, a pointwise convolution layer, an activation function, a global response normalization layer (GlobalResponse Normalization, GRN), and a stochastic dropout path layer. In the CCB, the present invention implements a three-dimensional version of GRN, and this method aims to constrain excessive feature divergence and improve training stability. It should be noted that the mathematical expression of GRN is as follows:
[0094] G x = Norm(F);
[0095]
[0096] F' = γ×(F + N x ) + β + F;
[0097] Wherein, the variables F and F' respectively represent the input and output feature maps. Norm(·) and Mean(·) respectively represent the operations of Euclidean distance normalization and mean calculation in the channel dimension of the feature map. γ and β respectively represent the first learnable parameter and the second learnable parameter, which are used for the weighted processing of the fused feature map. G x and N x are respectively the first intermediate variable and the second intermediate variable, and 1e -6 is set to prevent the mean from being 0.
[0098] As Figure 6 shown, Figure 6 shows the structural diagram of DB, including the downsampling layer, the position encoding layer, the normalization layer, the channel self-attention layer, the multi-layer perceptron layer, and the window self-attention layer. In DB, the channel attention uses the self-attention mechanism in the channel dimension of the feature map, while the window attention first divides the feature map into smaller 4×4×4 voxel sub-windows and performs the self-attention mechanism operation.
[0099] Step S40: Input the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain the final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
[0100] Specifically, calculate the brain age difference according to the actual age and the predicted brain age value, and generate the brain age attention weight through the fully connected layer and the sigmoid function of the brain age-guided attention module; input the AD-related features into the brain age-guided attention module for average pooling, multiply the result after average pooling by the brain age attention weight to obtain the final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
[0101] It can be seen that the present invention proposes a novel convolutional-transformer network guided by brain age information. This network framework mainly includes three components: a brain age prediction branch with frozen weights, an AD prediction branch, and a brain age information-guided attention module. Through this network framework, the complementary advantages of the convolutional and transformer network architectures can be fully utilized, and the robust classification of MRI images can be achieved by effectively using brain age knowledge.
[0102] Furthermore, as Figure 7As shown above, based on the above-mentioned MRI image classification method guided by brain age information, the present invention also correspondingly provides a system for classifying MRI images guided by brain age information. Among them, the system for classifying MRI images guided by brain age information includes:
[0103] A target image acquisition module 51, configured to acquire an MRI image and an actual age of a target object, preprocess the MRI image to obtain a target MRI image;
[0104] A target model construction module 52, configured to construct an MRI image classification model, train and test the MRI image classification model to obtain a target model. The target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer;
[0105] A dual-branch feature extraction module 53, configured to input the target MRI image into the brain age prediction branch of the target model for prediction, output a predicted brain age value, input the target MRI image into the AD prediction branch of the target model for prediction, and output AD-related features;
[0106] A feature fusion and classification module 54, configured to input the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain a final feature map, and obtain a classification result of the target MRI image after passing the final feature map through the fully connected layer.
[0107] Furthermore, as Figure 8 shown above, based on the above-mentioned MRI image classification method and system guided by brain age information, the present invention also correspondingly provides a terminal. The terminal includes a processor 10, a memory 20, and a display 30. Figure 8 Only some components of the terminal are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.
[0108] The memory 20 may be an internal storage unit of the terminal in some embodiments, such as the hard disk or memory of the terminal. The memory 20 may also be an external storage device of the terminal in some other embodiments, such as a plug-in hard disk equipped on the terminal, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 20 may also include both the internal storage unit of the terminal and the external storage device. The memory 20 is used to store the application software installed on the terminal and various types of data, such as the program code of the installed terminal, etc. The memory 20 may also be used to temporarily store the data that has been output or will be output. In one embodiment, a MRI image classification program 40 guided by brain age information is stored on the memory 20, and the MRI image classification program 40 guided by brain age information can be executed by the processor 10, so as to implement the MRI image classification method guided by brain age information in the present application.
[0109] The processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chips in some embodiments, and is used to run the program code stored in the memory 20 or process data, such as executing the MRI image classification method guided by brain age information, etc.
[0110] The display 30 may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 30 is used to display the information on the terminal and to display the visual user interface. The components of the terminal communicate with each other through the system bus.
[0111] In one embodiment, when the processor 10 executes the MRI image classification program 40 guided by brain age information stored in the memory 20, the following steps are implemented:
[0112] Obtain the MRI image and the actual age of the target object, preprocess the MRI image to obtain a target MRI image;
[0113] Construct a MRI image classification model, train and test the MRI image classification model to obtain a target model, and the target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer;
[0114] Input the target MRI image into the brain age prediction branch of the target model for prediction to output a predicted brain age value, and input the target MRI image into the AD prediction branch of the target model for prediction to output AD-related features;
[0115] Input the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain a final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
[0116] Among them, the brain age prediction branch includes a first PEB module and CRB modules in four stages;
[0117] The step of inputting the target MRI image into the brain age prediction branch of the target model for prediction to output a predicted brain age value specifically includes:
[0118] Input the target MRI image into the first PEB module for segmentation and encoding to obtain brain age encoded features;
[0119] Input the brain age encoded features into the CRB module in the first stage of the brain age prediction branch for Token fusion and channel fusion to obtain fused features, and the fused features include a first fused feature and a second fused feature;
[0120] Input the first fused feature and the second fused feature into the CRB module in the next stage to continue Token fusion and channel fusion until, after fusion by all CRB modules, a predicted brain age value is obtained.
[0121] Among them, the CRB module includes a first CRB architecture and a second CRB architecture;
[0122] The step of inputting the target MRI image into the CRB module in the first stage of the brain age prediction branch for Token fusion and channel fusion to obtain fused features specifically includes:
[0123] When the stride is 1, input the target MRI image into the first CRB architecture, perform downsampling on the target MRI image through parallel depth convolutional layers and residual layers, input the result of downsampling into a batch normalization layer for normalization processing, and integrate the normalized result through CCA to obtain a first Token fusion feature;
[0124] Output the first Token fusion feature to an inverted bottleneck structure for extraction and fusion of channel features to obtain a first fused feature, where the inverted bottleneck structure from left to right is: a pointwise convolutional layer, a first residual layer, a GeLU activation function, a pointwise convolutional layer, and a second residual layer;
[0125] When the step size is 2, input the target MRI image into the second CRB architecture, perform downsampling on the target MRI image through a depth convolutional layer, input the result of downsampling into a batch normalization layer for normalization processing, integrate the normalized result through CCA, double the dimension of the integrated result through a pointwise convolutional layer, and obtain a second Token fusion feature;
[0126] Output the second Token fusion feature to an inverted bottleneck structure for extraction and fusion of channel features, and obtain a second fusion feature.
[0127] Among them, the processing process of the CCA is as follows:
[0128] Input the input feature map into three different channels for average pooling respectively to obtain a first coordinate intermediate feature map, a second coordinate intermediate feature map, and a third coordinate intermediate feature map, rearrange the first coordinate intermediate feature map and the third coordinate intermediate feature map, and splice the rearranged first coordinate intermediate feature map, third coordinate intermediate feature map, and second coordinate intermediate feature map to obtain a first coordinate splicing result;
[0129] Input the first coordinate splicing result into a CBS module for processing to obtain a second coordinate splicing result, split the second coordinate splicing result to obtain three split feature maps, and pass the three split feature maps through a point convolutional layer and a sigmoid activation function to obtain a first coordinate feature, a second coordinate feature, and a third coordinate feature;
[0130] Input the input feature map into three different channels for average pooling, max pooling, and range pooling respectively to obtain a first channel intermediate feature map, a second channel intermediate feature map, and a third channel intermediate feature map, and splice the first channel intermediate feature map, second channel intermediate feature map, and third channel intermediate feature map to obtain a first channel splicing result;
[0131] Input the first channel splicing result into a CBS module for processing to obtain a second channel splicing result, and split the second coordinate splicing result to obtain a fourth channel intermediate feature map, a fifth channel intermediate feature map, and a sixth channel intermediate feature map;
[0132] Perform convolution on the intermediate feature map of the fourth channel and the intermediate feature map of the fifth channel through a point convolution layer, add the elements of the convolved intermediate feature map of the fourth channel and the convolved intermediate feature map of the fifth channel to obtain the intermediate feature map of the seventh channel, perform convolution on the intermediate feature map of the sixth channel through a point convolution layer, multiply the elements of the convolved intermediate feature map of the sixth channel and the intermediate feature map of the seventh channel to obtain the intermediate feature map of the eighth channel, and pass the intermediate feature map of the eighth channel through the sigmoid activation function to obtain the channel feature map;
[0133] Perform an element-wise outer product of the first coordinate feature, the second coordinate feature, the third coordinate feature, the channel feature map, and the input feature map to obtain the output feature map.
[0134] Among them, the CBS module includes a convolution layer, a normalization layer, and a swish activation function.
[0135] Among them, the AD prediction branch includes a second PEB module, two-stage CCB modules, and two-stage DB modules;
[0136] Input the target MRI image into the AD prediction branch of the target model for prediction, and output AD-related features, specifically including:
[0137] Input the target MRI image into the second PEB module for segmentation and encoding to obtain AD-encoded features;
[0138] Input the AD-encoded features into the CCB module for calculation to obtain CCB output features, and synchronously obtain the fused features from the CRB module, multiply the fused features by the corresponding learnable coefficients to obtain the first learning features, and add the first learning features and the CCB output features to obtain AD intermediate features;
[0139] Input the AD intermediate features into the DB module for calculation to obtain DB output features, and synchronously obtain the fused features from the CRB module, multiply the fused features by the corresponding learnable coefficients to obtain the second learning features, and add the second learning features and the DB output features to obtain AD-related features.
[0140] Among them, the step of inputting the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain the final feature map specifically includes:
[0141] Calculate the brain age difference based on the actual age and the predicted brain age value, and generate the brain age attention weight through the fully connected layer and the sigmoid function of the brain age-guided attention module;
[0142] Input the AD-related features into the brain age-guided attention module for average pooling, and multiply the result of the average pooling by the brain age attention weights to obtain the final feature map.
[0143] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an MRI image classification program guided by brain age information. When the MRI image classification program guided by brain age information is executed by a processor, the steps of the above-mentioned MRI image classification method guided by brain age information are implemented.
[0144] In summary, the present invention proposes an MRI image classification method, system, and terminal guided by brain age information. The method includes: obtaining an MRI image and the actual age of a target object, preprocessing the MRI image to obtain a target MRI image; constructing an MRI image classification model, training and testing the MRI image classification model to obtain a target model, and the target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer; inputting the target MRI image into the brain age prediction branch of the target model for prediction, outputting a predicted brain age value, inputting the target MRI image into the AD prediction branch of the target model for prediction, and outputting AD-related features; inputting the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain a final feature map, and passing the final feature map through the fully connected layer to obtain the classification result of the target MRI image. The present invention proposes a novel convolutional-transformer network guided by brain age information for integrating the features of two branches and giving the final AD classification result, making full use of the complementary advantages of the CNN and Transformer network architectures, and achieving a robust classification effect by effectively using brain age knowledge.
[0145] It should be noted that in this article, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or terminal including the element.
[0146] Of course, 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 computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.
[0147] It should be understood that the application of the present invention is not limited to the above examples. For those of ordinary skill in the art, improvements or transformations can be made according to the above description. All such improvements and transformations should fall within the protection scope of the appended claims of the present invention.
Claims
1. A method for classifying MRI images guided by brain age information, characterized in that, The described MRI image classification method guided by brain age information includes: Obtain the MRI image and actual age of the target object, preprocess the MRI image to obtain the target MRI image; Construct an MRI image classification model, train and test the MRI image classification model to obtain a target model, and the target model includes: a brain age prediction branch, an AD prediction branch, a brain age-guided attention module, and a fully connected layer; Input the target MRI image into the brain age prediction branch of the target model for prediction, output the predicted brain age value, and input the target MRI image into the AD prediction branch of the target model for prediction, output the AD-related features; Input the actual age, the predicted brain age value, and the AD-related features into the brain age-guided attention module of the target model for integration to obtain the final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
2. The MRI image classification method guided by brain age information according to claim 1, wherein The brain age prediction branch includes a first PEB module and four stages of CRB modules; The step of inputting the target MRI image into the brain age prediction branch of the target model for prediction and outputting the predicted brain age value specifically includes: Input the target MRI image into the first PEB module for segmentation and encoding to obtain brain age encoded features; Input the brain age encoded features into the CRB module of the first stage in the brain age prediction branch for Token fusion and channel fusion to obtain fusion features, and the fusion features include a first fusion feature and a second fusion feature; Input the first fusion feature and the second fusion feature into the CRB module of the next stage to continue performing Token fusion and channel fusion until after all CRB modules are fused, and obtain the predicted brain age value.
3. The MRI image classification method guided by brain age information according to claim 2, wherein The CRB module includes a first CRB architecture and a second CRB architecture; The step of inputting the target MRI image into the CRB module of the first stage in the brain age prediction branch for Token fusion and channel fusion to obtain fusion features specifically includes: When the stride is 1, input the target MRI image into the first CRB architecture, perform downsampling on the target MRI image through parallel depth convolutional layers and residual layers, input the downsampled result into a batch normalization layer for normalization processing, and integrate the normalized result through CCA to obtain a first Token fusion feature; Output the first Token fusion feature to an inverted bottleneck structure for extraction and fusion of channel features to obtain a first fusion feature, where the inverted bottleneck structure is, from left to right: a pointwise convolutional layer, a first residual layer, a GeLU activation function, a pointwise convolutional layer, and a second residual layer; When the step size is 2, input the target MRI image into the second CRB architecture, perform downsampling on the target MRI image through the depth convolution layer, input the result of downsampling into the batch normalization layer for normalization processing, integrate the normalized result through CCA, and double the dimension of the integrated result through the pointwise convolution layer to obtain the second Token fusion feature; Output the second Token fusion feature to the inverted bottleneck structure for extraction and fusion of channel features to obtain the second fusion feature.
4. The MRI image classification method guided by brain age information according to claim 3, characterized in that The processing process of the CCA is as follows: Input the input feature map into three different channels for average pooling respectively to obtain the first coordinate intermediate feature map, the second coordinate intermediate feature map, and the third coordinate intermediate feature map, rearrange the first coordinate intermediate feature map and the third coordinate intermediate feature map, and splice the rearranged first coordinate intermediate feature map, the third coordinate intermediate feature map, and the second coordinate intermediate feature map to obtain the first coordinate splicing result; Input the first coordinate splicing result into the CBS module for processing to obtain the second coordinate splicing result, split the second coordinate splicing result to obtain three split feature maps, and pass the three split feature maps through the point convolution layer and the sigmoid activation function to obtain the first coordinate feature, the second coordinate feature, and the third coordinate feature; Input the input feature map into three different channels for average pooling, max pooling, and range pooling respectively to obtain the first channel intermediate feature map, the second channel intermediate feature map, and the third channel intermediate feature map, and splice the first channel intermediate feature map, the second channel intermediate feature map, and the third channel intermediate feature map to obtain the first channel splicing result; Input the first channel splicing result into the CBS module for processing to obtain the second channel splicing result, and split the second coordinate splicing result to obtain the fourth channel intermediate feature map, the fifth channel intermediate feature map, and the sixth channel intermediate feature map; Convolve the fourth channel intermediate feature map and the fifth channel intermediate feature map through the point convolution layer, add the elements of the convolved fourth channel intermediate feature map and the convolved fifth channel intermediate feature map to obtain the seventh channel intermediate feature map, convolve the sixth channel intermediate feature map through the point convolution layer, multiply the elements of the convolved sixth channel intermediate feature map and the seventh channel intermediate feature map to obtain the eighth channel intermediate feature map, and pass the eighth channel intermediate feature map through the sigmoid activation function to obtain the channel feature map; Perform element-wise outer product on the first coordinate feature, the second coordinate feature, the third coordinate feature, the channel feature map, and the input feature map to obtain the output feature map.
5. The MRI image classification method guided by brain age information according to claim 4, wherein The CBS module includes a convolution layer, a normalization layer, and a swish activation function.
6. The MRI image classification method guided by brain age information according to claim 3, wherein The AD prediction branch includes a second PEB module, two-stage CCB modules, and two-stage DB modules; Input the target MRI image into the AD prediction branch of the target model for prediction and output AD-related features, specifically including: Input the target MRI image into the second PEB module for segmentation and encoding to obtain AD encoded features; Input the AD encoded features into the CCB module for calculation to obtain CCB output features, and synchronously obtain the fused features from the CRB module. Multiply the fused features by the corresponding learnable coefficients to obtain the first learned features, and add the first learned features and the CCB output features to obtain AD intermediate features; Input the AD intermediate features into the DB module for calculation to obtain DB output features, and synchronously obtain the fused features from the CRB module. Multiply the fused features by the corresponding learnable coefficients to obtain the second learned features, and add the second learned features and the DB output features to obtain AD related features.
7. The MRI image classification method guided by brain age information according to claim 1, wherein Integrating the actual age, the predicted brain age value, and the AD related features into the brain age guided attention module of the target model to obtain the final feature map specifically includes: Calculate the brain age difference based on the actual age and the predicted brain age value, and generate brain age attention weights through the fully connected layer and the sigmoid function of the brain age guided attention module; Input the AD related features into the brain age guided attention module for average pooling, and multiply the result of the average pooling by the brain age attention weights to obtain the final feature map.
8. An MRI image classification system guided by brain age information, characterized in that, The MRI image classification system guided by brain age information includes: A target image acquisition module, configured to acquire the MRI image and the actual age of a target object, preprocess the MRI image to obtain a target MRI image; A target model construction module, configured to construct an MRI image classification model, train and test the MRI image classification model to obtain a target model, where the target model includes: a brain age prediction branch, an AD prediction branch, a brain age guided attention module, and a fully connected layer; A dual-branch feature extraction module, configured to input the target MRI image into the brain age prediction branch of the target model for prediction to output a predicted brain age value, and input the target MRI image into the AD prediction branch of the target model for prediction to output AD related features; A feature fusion and classification module, configured to integrate the actual age, the predicted brain age value, and the AD related features into the brain age guided attention module of the target model to obtain a final feature map, and obtain the classification result of the target MRI image after passing the final feature map through the fully connected layer.
9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a brain age information-guided MRI image classification program stored on the memory and executable on the processor. When the brain age information-guided MRI image classification program is executed by the processor, it implements the steps of the brain age information-guided MRI image classification method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an MRI image classification program guided by brain age information. When the MRI image classification program guided by brain age information is executed by a processor, the steps of the MRI image classification method guided by brain age information according to any one of claims 1-7 are implemented.