Individual brain mapping method and device based on brain mapping fusion model

Abstract

The application relates to an individual brain mapping method and device based on a brain mapping fusion model. The method comprises the following steps: receiving at least two groups of subject data for training of a brain mapping fusion model, extracting data features of the subject data, and pre-processing the subject data to obtain an adjacency matrix; inputting the data features and the adjacency matrix into an initial brain mapping fusion model based on a graph convolution model and a label propagation model, obtaining brain mapping prediction values output by the initial brain mapping fusion model, training the initial brain mapping fusion model based on the graph convolution and the label propagation algorithm according to the brain mapping prediction values, and obtaining a trained brain mapping fusion model; and inputting to-be-detected data into the trained brain mapping fusion model to obtain an individual brain map. The method can combine the graph convolution algorithm and the label propagation algorithm to process the subject data, and improve the drawing speed and precision of the individual brain map.

Description

Technical Field

[0001] This application relates to the fields of medical imaging and deep learning technology, and in particular to a method and apparatus for drawing individual brain maps based on a brain map fusion model. Background Technology

[0002] The regional division of the cerebral cortex, i.e., brain mapping, is an important part of exploring brain function and structure. With the development of magnetic resonance imaging (MRI) technology, resting-state functional magnetic resonance imaging (fMRI) and diffusion magnetic resonance imaging (DMRI) are increasingly being used in brain mapping research, especially in brain mapping studies that can reflect individual differences.

[0003] However, there are still many challenges in the current technology of individual brain mapping based on magnetic resonance. For example, in the process of processing diffuse magnetic resonance signals using graph convolution plus Chebyshev convolution kernel, the adjacency matrix needs to be processed by Chebyshev convolution kernel multiple times, and the order needs to be determined by parameter tuning, etc. The above-mentioned shortcomings have a significant impact on the speed and accuracy of brain mapping, hindering the development of many studies such as craniocerebral surgery and individual brain development tracking. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and apparatus for individual brain mapping based on a brain mapping fusion model that can improve the accuracy and speed of individual brain mapping, thereby addressing the aforementioned technical problems.

[0005] Firstly, this application provides a method for drawing individual brain maps based on a brain map fusion model. The method includes:

[0006] Receive at least two sets of subject data for training a brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix;

[0007] The data features and the adjacency matrix are input into the constructed initial brain map fusion model based on graph convolution and label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithm is trained according to the brain map prediction value to obtain the trained brain map fusion model.

[0008] The test data is input into the trained brain map fusion model to obtain an individual brain map.

[0009] In one embodiment, receiving at least two sets of subject data for training a brain mapping fusion model, extracting data features from the subject data, and preprocessing the subject data to obtain an adjacency matrix includes:

[0010] Receive at least two sets of subject data for training the brain mapping fusion model, each set of subject data including T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion-weighted fMRI data;

[0011] Data feature extraction processing is performed on the T1-weighted data and the resting-state functional magnetic resonance data to obtain the adjacency matrix corresponding to the T1-weighted data and the adjacency matrix corresponding to the resting-state functional magnetic resonance data;

[0012] Fiber bundle features are extracted from the diffusion-weighted magnetic resonance data.

[0013] In one embodiment, the step of performing data feature extraction processing on the T1-weighted data and the resting-state functional magnetic resonance imaging (fMRI) data to obtain the adjacency matrix corresponding to the T1-weighted data and the adjacency matrix corresponding to the resting-state fMRI data includes:

[0014] The cortical surface structure is extracted from the T1 weighted data, the cortical surface structure is mapped to a standard template, and the cortical vertex adjacency matrix is ​​calculated based on the relationship between the triangular facet edges of the mapped cortical surface structure.

[0015] Vertex time series are extracted from the resting-state functional magnetic resonance data. The Euclidean distance between any two vertices in the vertex time series is calculated. The dot product of the connection weights between the vertices and the Euclidean distances is calculated. The dot product is minimized iteratively to obtain a weighted adjacency matrix.

[0016] In one embodiment, extracting fiber bundle features from the diffusion-weighted magnetic resonance data includes:

[0017] The initial fiber bundle features of the diffusion-weighted magnetic resonance data are extracted, the number of voxels of the initial fiber bundle features is traversed, and the initial fiber bundle features are filled according to the number of longest voxels to obtain the fiber bundle features.

[0018] In one embodiment, the data features and the adjacency matrix are input into an existing initial brain map fusion model based on a graph convolution model and a label propagation model to obtain a brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model is then trained based on the brain map prediction value to obtain a trained brain map fusion model, including:

[0019] The adjacency matrix corresponding to the T1 weighted data is normalized to obtain a normalized adjacency matrix, and the fiber bundle features are dimension-reduced to obtain dimension-reduced fiber bundle features.

[0020] Construct a first graph convolutional model, and input the normalized adjacency matrix and the reduced fiber bundle features into the first graph convolutional model to obtain the first brain map prediction value;

[0021] A label propagation model is constructed, and the normalized adjacency matrix and the preset group map labels are input into the label propagation model to obtain the second brain map prediction value;

[0022] A second graph convolutional model is constructed. The preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data are input into the second graph convolutional model to obtain the third brain map prediction value.

[0023] The initial brain map fusion model is trained based on the first brain map prediction value, the second brain map prediction value, and the third brain map prediction value to obtain the trained brain map fusion model.

[0024] In one embodiment, constructing the second graph convolutional model involves inputting a preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data into the second graph convolutional model to obtain the third brain map prediction value, including:

[0025] Construct a second graph convolutional model that includes two graph convolutional layers;

[0026] The data features corresponding to the resting-state functional magnetic resonance data and the adjacency matrix corresponding to the resting-state functional magnetic resonance data are input into the second graph convolution model to obtain the output result of the second graph convolution model;

[0027] The output is multiplied by a preset prior mask to obtain the predicted value of the third brain map.

[0028] In one embodiment, the initial brain map fusion model is trained based on the first brain map prediction value, the second brain map prediction value, and the third brain map prediction value to obtain a trained brain map fusion model, including:

[0029] The first brain map prediction value and the third brain map prediction value are stacked to obtain a stacked result. The stacked result is then processed into a linear layer to obtain a fourth brain map prediction value.

[0030] A first loss function is obtained by calculating the cross-entropy between the predicted value of the fourth brain map and the preset group map; a second loss function is obtained by calculating the cross-entropy between the predicted value of the second brain map and the preset group map.

[0031] The first loss function and the second loss function are weighted and calculated to obtain the final loss function;

[0032] Iterate the final loss function until it minimizes convergence to obtain the trained brain map fusion model.

[0033] Secondly, this application also provides an individual brain mapping device based on a brain mapping fusion model. The device includes:

[0034] The feature extraction module is used to receive at least two sets of subject data for training the brain map fusion model, extract the data features of the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0035] The training model module is used to input the data features and the adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithm is trained according to the brain map prediction value to obtain the trained brain map fusion model.

[0036] The individual brain mapping module is used to input the test data into the trained brain mapping fusion model to obtain the individual brain map.

[0037] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0038] Receive at least two sets of subject data for training a brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix;

[0039] The data features and the adjacency matrix are input into the constructed initial brain map fusion model based on graph convolution and label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithm is trained according to the brain map prediction value to obtain the trained brain map fusion model.

[0040] The test data is input into the trained brain map fusion model to obtain an individual brain map.

[0041] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0042] Receive at least two sets of subject data for training a brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix;

[0043] The data features and the adjacency matrix are input into the constructed initial brain map fusion model based on graph convolution and label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithm is trained according to the brain map prediction value to obtain the trained brain map fusion model.

[0044] The test data is input into the trained brain map fusion model to obtain an individual brain map.

[0045] The aforementioned method and apparatus for drawing individual brain maps based on a brain map fusion model inputs the data features and adjacency matrix of at least two sets of subject data into an initial brain map fusion model based on a graph convolution model and a label propagation model. This yields the brain map prediction value output by the initial brain map fusion model. The model is then trained based on the brain map prediction value to obtain the trained brain map fusion model. By combining the graph convolution algorithm and the label propagation algorithm to process the subject data, it eliminates the need for iterative calculation of Chebyshev convolution kernels, thereby improving the speed and accuracy of model training and individual brain map drawing. Attached Figure Description

[0046] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0047] Figure 1 This is an application environment diagram of an individual brain mapping method based on a brain mapping fusion model in one embodiment;

[0048] Figure 2 This is a flowchart illustrating an individual brain mapping method based on a brain mapping fusion model in one embodiment.

[0049] Figure 3 This is a flowchart illustrating an individual brain mapping method based on a brain mapping fusion model in a preferred embodiment.

[0050] Figure 4 This is a data processing block diagram corresponding to a preferred embodiment;

[0051] Figure 5 This is a structural block diagram of an individual brain mapping device based on a brain mapping fusion model in one embodiment;

[0052] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0054] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0055] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0056] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0057] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the individual brain mapping method based on a brain mapping fusion model in this embodiment. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0058] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the individual brain mapping method based on a brain mapping fusion model in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0059] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0060] This embodiment provides a method for drawing individual brain maps based on a brain map fusion model. Figure 2 This is a flowchart of the individual brain mapping method based on a brain mapping fusion model in this embodiment, as shown below. Figure 2 As shown, the process includes the following steps:

[0061] Step S210: Receive at least two sets of subject data for training the brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0062] The subject data includes at least two sets of subject data from subjects who underwent experiments or tests. Each subject data set includes T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion magnetic resonance imaging (DMI) data. The process of extracting data features from the subject data and preprocessing the subject data to obtain the adjacency matrix specifically includes: (1) T1-weighted data uses FreeSurfer to extract the cortical surface structure, maps it onto a standard template, and calculates the cortical vertex adjacency matrix based on the edge relationships of the triangular facets of the mapped cortical structure; (2) Resting-state fMRI data uses fmriprep containing ICA-AROMA to extract the vertex time series, and further obtains the weighted adjacency matrix of the smoothed vertex time series by minimizing the total variation of the graph signal; (3) DMI data uses FSL to obtain fiber bundle features with the number of vertices × the number of voxels. The number of vertices × the number of voxels indicates that the dimension of the obtained fiber bundle features is (number of vertices, number of voxels).

[0063] Step S220: Input the data features and adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation models to obtain the brain map prediction value output by the initial brain map fusion model. Train the initial brain map fusion model based on graph convolution and label propagation algorithms according to the brain map prediction value to obtain the trained brain map fusion model.

[0064] In this training process, the data features of each subject's data are used as input data for each round of training. These data features include vertex time series and fiber bundle features. The adjacency matrix includes the cortical vertex adjacency matrix corresponding to the T1-weighted data and the weighted adjacency matrix corresponding to the resting-state functional magnetic resonance imaging (fMRI) data. The connection weights of the cortical vertex adjacency matrix are mapped to the [0,1] distribution interval using softmax to obtain a normalized adjacency matrix, which also serves as a trainable variable for the label propagation algorithm in the fusion model. The fiber bundle features are input into the embedding layer for dimensionality reduction, resulting in dimensionality-reduced fiber bundle features. For example, this dimensionality-reduced fiber bundle feature matrix is ​​a fiber bundle feature matrix of vertices × 72.

[0065] The initial brain map fusion models include graph convolution models, label propagation models, and graph convolution models with masked inputs.

[0066] A graph convolution model is constructed, and the reduced fiber bundle features and the normalized adjacency matrix are input into the graph convolution model. The fiber bundle features are classified through graph convolution to obtain the first brain map prediction value.

[0067] A label propagation model is constructed by inputting the normalized adjacency matrix and the group graph labels into the label propagation model to obtain a set of second brain graph propagation values.

[0068] A prior mask and a graph convolutional model with masked input are constructed using vertex time series and group graph labels. The vertex time series, prior mask, and weighted adjacency matrix are fed into the graph convolutional model with masked input, and the predicted value of the third brain map is obtained through graph convolutional classification.

[0069] The predicted values ​​of the first and third brain maps are stacked together and passed through a linear layer to obtain the predicted value of the fourth brain map. Then, the cross-entropy between the predicted values ​​and the group map is calculated to obtain the first loss function. The propagated value of the second brain map is directly cross-entropy with the group map to obtain the second loss function. The second loss function is multiplied by a coefficient and added to the first loss function to obtain the final loss function. The final loss function is iterated until convergence, and the brain map fusion model training is completed.

[0070] Step S230: Input the data to be tested into the trained brain map fusion model to obtain an individual brain map.

[0071] The received brain data of the subject to be predicted is preprocessed to obtain test data including the adjacency matrix of the cortical vertices to be tested, the time series of the vertices to be tested, the weighted adjacency matrix to be tested, and the fiber tract features to be tested. This test data is then input into a trained brain mapping fusion model to obtain predicted values ​​for individual diffuse magnetic resonance imaging (DMRI) brain atlases, functional magnetic resonance imaging (fMRI) brain atlases, and fusion model predictions. An individual brain atlas is obtained based on the fusion model predictions. Additionally, individual DMRI brain atlases and functional MRI brain atlases can be obtained from the predicted values ​​of the individual DMRI brain atlases, serving as supplementary references for physicians. DMRI brain atlases can more accurately reflect brain structure, while fMRI brain atlases can more accurately reflect the status of active brain regions.

[0072] Furthermore, individual brain atlases, diffusion magnetic resonance imaging (DMRI) individual brain atlases, and functional magnetic resonance imaging (fMRI) individual brain atlases are mapped onto the cortical surface through the individual brain atlas display module and displayed in 3D.

[0073] In related technologies, on the one hand, fitting individual brain maps using subject data from a single test subject is not only time-consuming due to iterative solutions each time, but also prone to overfitting, resulting in poor prediction quality. On the other hand, in processing diffuse magnetic resonance signals using graph convolution plus Chebyshev convolution kernels, processing the adjacency matrix with Chebyshev convolution kernels requires multiple iterations, and the order needs to be determined through parameter tuning, leading to low prediction efficiency and accuracy.

[0074] In the aforementioned method for individual brain mapping based on a brain mapping fusion model, the data features and adjacency matrix of at least two sets of subject data are input into an initial brain mapping fusion model based on a graph convolution model and a label propagation model. This initial model is then trained to obtain a brain mapping fusion model that can be reused for prediction data from different individuals. Combining graph convolution and label propagation algorithms to process subject data eliminates the need for iterative calculations of Chebyshev convolution kernels, improving the speed and accuracy of model training and individual brain mapping.

[0075] In one embodiment, based on step S210 above, at least two sets of subject data for training the brain mapping fusion model are received, data features of the subject data are extracted, and the subject data is preprocessed to obtain an adjacency matrix. Specifically, this may include the following steps:

[0076] Step S211: Receive at least two sets of subject data for training the brain mapping fusion model, each set of subject data including T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion-weighted fMRI data.

[0077] For example, the Longitudinal Multimodal Imaging (SLIM) Brain Database of ** University was downloaded. This database contains long-term test-retest samples of young, healthy adults in Southwest China. After filtering out some diffusion-weighted data with defects, 576 subjects were obtained. Each subject's data included at least one set of T1-weighted data, diffusion-weighted MRI data, and resting-state functional MRI data. The first 400 sets of data were selected as subjects for training the brain mapping fusion model.

[0078] Step S212: Perform data feature extraction processing on the T1-weighted data and resting-state functional magnetic resonance data to obtain the adjacency matrix corresponding to the T1-weighted data and the adjacency matrix corresponding to the resting-state functional magnetic resonance data.

[0079] In one embodiment, step S212 includes:

[0080] Step S2121: Extract the cortical surface structure from the T1 weighted data, map the cortical surface structure onto the standard template, and calculate the cortical vertex adjacency matrix based on the relationship between the triangular facet edges of the mapped cortical surface structure.

[0081] Step S2122: Extract vertex time series from resting-state functional magnetic resonance data, calculate the Euclidean distance between any two vertices in the vertex time series, calculate the dot product of the connection weights between vertices and the Euclidean distances, perform iterative calculation to minimize the dot product sum, and obtain a weighted adjacency matrix.

[0082] For example, after arranging the subject data used for training in bids format, the fmriprep tool can be used to process T1-weighted data from multiple subjects in parallel for resting-state functional MRI data preprocessing. The output-space parameter is selected based on the resolution chosen during processing, and the use-aroma parameter is selected to perform ICA-AROMA, used to eliminate head motion effects in resting-state functional MRI. After the fmriprep command is completed, vertex time-series features are extracted from the resting-state functional MRI data, with a size of 10242×2^10, where 10242 is the number of vertices and 2^10 is the time length. The T1-weighted data is then stored in the freesurfer folder.

[0083] The process of generating the adjacency matrix for T1-weighted data includes: using the commands of the freesurfer tool, locating the surf folder under each subject's folder in the freesurfer folder, obtaining the spherical, white matter, and soft membrane files for each subject, and using surf2surf to convert these files to the required resolution. Options include fsaverage3 (642 nodes), fsaverage4 (2562 nodes), fsaverage5 (10242 nodes), fsaverage6 (40962 nodes), and fsaverage7 (163842 nodes). Here, fsaverage5 (10k) is used as an example. Then, mris_convert converts these files into surface files in surf.gii format that can be read by Python. Python is used to read the white matter surface files, and the adjacency matrix is ​​obtained based on the adjacency relationships of the triangular facet edges, with a shape of 10242×10242.

[0084] The calculation process of the weighted adjacency matrix specifically includes: using the gsp_distanz function of the gspbox tool to first obtain the Euclidean distance Z between any two vertices in the vertex time series:

[0085] Z[i,j]=||X[i,:]-X[j,:]|| 2 ;

[0086] Where X is the vertex time series, i, j are vertex indices, ||`|| 2 The colon (:) indicates the Euclidean distance, and the colon (:) indicates the data in all columns.

[0087] Then, using `gsp_learn_graph_log_degrees` iteratively minimize the sum of the dot product of the connection weight W between any two vertices and their Euclidean distance Z, we obtain the weighted adjacency matrix of the smoothed graph signal:

[0088]

[0089] In this context, the first term ".*" represents dot product, and the next two terms are regularization factors. Here, 'a' controls the value of W, with larger 'a' values ​​resulting in larger W values, and 'b' controls the sparsity of W, with smaller 'b' values ​​resulting in sparser W. '||`||F' represents Frobenius regularization.

[0090] Step S2122 above updates the weights of the adjacency matrix by minimizing the total variation of the graph signal, thereby smoothing the signal and improving the robustness of the trained brain map fusion model.

[0091] Step S213: Extract fiber bundle features from diffusion-weighted magnetic resonance data.

[0092] In one embodiment, step S213 includes:

[0093] The initial fiber bundle features are extracted from the diffusion-weighted magnetic resonance data. The number of voxels in the initial fiber bundle features is traversed, and the initial fiber bundle features are filled according to the number of longest voxels to obtain the fiber bundle features.

[0094] For example, referring to the pipeline provided by the FSL tool for processing diffusion-weighted magnetic resonance data, when using probtrackx2_gpu to trace fiber bundles, a white matter surface file is required as the starting point for tracing. After tracing, fiber bundle features are obtained in the form of vertex number × voxel number. The number of voxels varies from person to person. The initial fiber bundle features are padded according to the number of longest voxels to obtain fiber bundle features with dimensions of vertex number × longest voxel number. For example, the fiber bundle features are padded with zero values ​​or truncated to a length of 67076, resulting in a fiber bundle feature size of 10242 × 67076.

[0095] Steps S211 to S213 above train the initial brain map fusion model using subject data from multiple subjects. Compared to a prediction model trained using subject data from a single subject, which can only predict the same individual, the brain map fusion model in this embodiment can be reused for different test subjects, adapting to more test subjects, improving the speed of drawing individual brain maps, and avoiding overfitting.

[0096] In one embodiment, based on step S220 above, data features and adjacency matrix are input into the constructed initial brain map fusion model based on graph convolution and label propagation models to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithms is then trained based on the brain map prediction value to obtain the trained brain map fusion model. Specifically, this may include the following steps:

[0097] Step S221: Normalize the adjacency matrix corresponding to the weighted data of T1 to obtain the normalized adjacency matrix, and reduce the dimensionality of the fiber bundle features to obtain the dimensionality-reduced fiber bundle features.

[0098] In this process, the connection weights of the adjacency matrix are mapped to the distribution interval [0,1] using softmax to obtain a normalized adjacency matrix, which also serves as a trainable variable for the label propagation algorithm. The fiber bundle features are input into the embedding layer for dimensionality reduction, and the output fiber bundle features are 72 times the number of vertices.

[0099] Step S222: Construct the first graph convolution model by inputting the normalized adjacency matrix and the reduced fiber bundle features into the first graph convolution model to obtain the first brain map prediction value.

[0100] The first graph convolutional model consists of five graph convolutional layers. The first layer has an input size of 72 and an output size of 64. The second to fourth layers all have an input and output size of 64, and the last layer has an input size of 64 and an output size of 105. The first brain map prediction value obtained after passing through the first graph convolutional model is the probability of each label on the cortical vertex, with a value equal to the number of vertices multiplied by the number of labels, i.e., 10242 × 10.

[0101] Step S223: Construct a label propagation model by inputting the normalized adjacency matrix and the preset group map labels into the label propagation model to obtain the second brain map prediction value.

[0102] The label propagation model allows the group graph labels to propagate through a normalized adjacency matrix with trainable edge weights, resulting in a set of propagated label values, which are the predicted values ​​of the second brain graph. The shape is the number of vertices × 1, i.e., 10242 × 1.

[0103] Step S224: Construct a second graph convolutional model by inputting the preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data into the second graph convolutional model to obtain the third brain map prediction value.

[0104] In one embodiment, step S224 above includes:

[0105] Step S2241: Construct a second graph convolutional model that includes two graph convolutional layers.

[0106] The second graph convolution model is a graph convolution model with masked input. A dropout layer is added before each graph convolution layer to prevent overfitting. The input size of the first graph convolution layer is the length of the time series, 2^10, and the output is 64. The input size of the second graph convolution layer is 64, and the output is set to the number of map labels, 105.

[0107] Step S2242: Input the data features corresponding to the resting-state functional magnetic resonance data and the adjacency matrix corresponding to the resting-state functional magnetic resonance data into the second graph convolution model to obtain the output result of the second graph convolution model.

[0108] Specifically, the resting-state functional magnetic resonance (fMRI) data can be processed through step S2122 above to obtain the data features corresponding to the resting-state fMRI data, namely the vertex time series; and to obtain the adjacency matrix corresponding to the resting-state fMRI data, namely the weighted adjacency matrix. The vertex time series and the weighted adjacency matrix are then input into the second graph convolution model to obtain the output result of the second graph convolution model.

[0109] Step S2243: Multiply the output result with the preset prior mask to obtain the predicted value of the third brain map.

[0110] The preset prior mask can be constructed using the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, namely the vertex time series and the group map labels. Specifically, the prior mask is a set of Boolean values ​​with the same shape as the group map, and its construction process is as follows: Match the vertex indices corresponding to the group map labels with the vertices in the vertex time series, calculate the average value of the matched vertices in the time series to obtain the label time series (time length × number of labels); calculate the Pearson correlation of the time dimension of the vertex time series and the label time series to obtain the vertex label probability matrix; select the group map label with the highest probability according to the vertex label probability matrix to obtain the label prediction value; construct the prior mask based on the label prediction value and the group map label. Specifically, compare the label prediction value and the group map label, setting the mask value to 1 for the same part and 0 for the different parts.

[0111] The predicted value of the third brain map is the same as that of the first brain map, which is the number of vertices × the number of labels, i.e., 10242 × 105.

[0112] Steps S2241 to S2243 above use a masked graph convolution algorithm to process resting-state functional magnetic resonance data, which can resist unstable fluctuations in the signal and improve the accuracy of the model.

[0113] Step S225: Train the initial brain map fusion model based on the prediction values ​​of the first brain map, the second brain map, and the third brain map to obtain the trained brain map fusion model.

[0114] In one embodiment, step S225 includes:

[0115] Step S2251: Stack the predicted values ​​of the first brain map and the third brain map to obtain a stacked result, and then send the stacked result into a linear layer for processing to obtain the predicted value of the fourth brain map.

[0116] Step S2252: Calculate the cross-entropy between the predicted value of the fourth brain map and the preset group map to obtain the first loss function.

[0117] The predicted value of the fourth brain map is the number of vertices multiplied by the number of labels, i.e., 10242 × 10⁵. The cross-entropy of the predicted value of the fourth brain map and the labels of the group map after one-hot encoding is calculated to obtain the first loss function. The formula for the first loss function, Loss1, is as follows:

[0118]

[0119] Where p is the predicted value of the first brain map after passing through the linear layer, y is the group map label value, and i is the corresponding cortical vertex index, with values ​​ranging from positive integers.

[0120] Step S2253: Calculate the cross-entropy between the predicted value of the second brain map and the preset group map to obtain the second loss function.

[0121] Find the cross-entropy (Loss2) of the loss function between the predicted values ​​from the label propagation algorithm and the group map spectrum.

[0122]

[0123] Where p is the predicted value of the second brain map, y is the group map label value, and i is the corresponding cortical vertex index, with values ​​ranging from positive integers.

[0124] Step S2254: The first loss function and the second loss function are weighted and summed to obtain the final loss function. The formula for calculating the final loss function (Loss) is as follows:

[0125] Loss = Loss1 + α × Loss2;

[0126] The coefficient α is given as a parameter.

[0127] Step S2255: Iterate the final loss function until it minimizes and converges to obtain the trained brain map fusion model.

[0128] After obtaining the final loss function, backpropagation is performed until the model converges, and the training weights are saved.

[0129] In this embodiment, the initial brain mapping fusion model is input using T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion magnetic resonance imaging (DMRI) data from multiple subjects. Minimizing the total variation of the graph signal and masked graph convolution are used to process the resting-state fMRI data to resist signal instability. The advantages of both graph convolution and label propagation algorithms are combined to process DMRI and T1-weighted signals, eliminating the need for iterative calculations of Chebyshev convolution kernels and improving model speed. Training the model with MRI data from multiple subjects enhances its robustness.

[0130] The present embodiment will now be described and illustrated through preferred embodiments.

[0131] Figure 3 This is a flowchart of an individual brain mapping method based on a brain mapping fusion model according to a preferred embodiment of the present invention.

[0132] Step S301: Receive at least two sets of subject data for training the brain mapping fusion model, each set of subject data including T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion-weighted fMRI data.

[0133] Step S302: Extract the cortical surface structure from the T1 weighted data, map the cortical surface structure to a standard template, and calculate the cortical vertex adjacency matrix based on the relationship between the triangular facet edges of the mapped cortical surface structure; extract the vertex time series from the resting-state functional magnetic resonance imaging data, calculate the Euclidean distance between any two vertices in the vertex time series, calculate the sum of the dot product of the connection weights between vertices and the Euclidean distances, and perform iterative calculation to minimize the sum of the dot products to obtain the weighted adjacency matrix.

[0134] Step S303: Normalize the adjacency matrix of cortical vertices to obtain a normalized adjacency matrix, and reduce the dimensionality of fiber bundle features to obtain dimensionality-reduced fiber bundle features.

[0135] Step S304: Construct a first graph convolutional model by inputting the normalized adjacency matrix and the reduced fiber bundle features into the first graph convolutional model to obtain the first brain map prediction value; construct a label propagation model by inputting the normalized adjacency matrix and the preset group map labels into the label propagation model to obtain the second brain map prediction value; construct a second graph convolutional model by inputting the preset prior mask, vertex time series and weighted adjacency matrix into the second graph convolutional model to obtain the third brain map prediction value.

[0136] Step S305: Train the initial brain map fusion model based on the predicted values ​​of the first brain map, the second brain map, and the third brain map to obtain the trained brain map fusion model.

[0137] Step S306: Input the data to be tested into the trained brain map fusion model to obtain an individual brain map.

[0138] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0139] Figure 4 This is a data processing block diagram corresponding to a preferred embodiment of the present invention. For example... Figure 4 As shown, the specific content of this data frame diagram is as follows:

[0140] When creating a personalized brain map of the test subject, the test data 41 is input into the brain map fusion model 42. The brain map fusion model 42 predicts the individual brain map 43 of the current subject after predicting the input data.

[0141] In addition, before using the brain mapping fusion model 42 for prediction, the brain mapping fusion model 42 needs to be trained using the test data to obtain a brain mapping fusion model 42 that can be used for accurate prediction.

[0142] Among them, the training data 44 comes from at least two sets of T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion-weighted magnetic resonance imaging (DMI) data corresponding to the subjects; the test data 41 is a set of T1-weighted data, resting-state fMRI data, and DMI data corresponding to the subjects.

[0143] Based on the same inventive concept, this embodiment also provides an individual brain mapping device 50 based on a brain mapping fusion model. This system is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. The terms "module," "unit," "subunit," etc., used below refer to combinations of software and / or hardware that achieve a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0144] In one embodiment, such as Figure 5 As shown, an individual brain atlas mapping device 50 based on a brain atlas fusion model is provided, including: a feature extraction module 51, a training model module 52, and an individual brain atlas mapping module 53, wherein:

[0145] The feature extraction module 51 is used to receive at least two sets of subject data for training the brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0146] The subject data includes subject data from at least two subjects who underwent the experiment or test. Each subject data includes T1-weighted data, resting-state functional magnetic resonance imaging (fMRI) data, and diffusion magnetic resonance imaging (DMI) data. The process of extracting data features from the subject data and preprocessing the subject data to obtain the adjacency matrix specifically includes: (1) T1-weighted data uses FreeSurfer to extract the cortical surface structure, maps it onto a standard template, and calculates the cortical vertex adjacency matrix based on the edge relationships of the triangular facets of the mapped cortical structure; (2) Resting-state fMRI data uses fmriprep containing ICA-AROMA to extract the vertex time series, and further obtains the weighted adjacency matrix of the smoothed vertex time series by minimizing the total variation of the graph signal; (3) DMI data uses FSL to obtain fiber bundle features of the number of vertices × the number of voxels. The number of vertices × the number of voxels indicates that the dimension of the obtained fiber bundle features is (number of vertices, number of voxels).

[0147] The training model module 52 is used to input data features and adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model based on graph convolution and label propagation algorithm is trained according to the brain map prediction value to obtain the trained brain map fusion model.

[0148] In this training process, the data features of each subject's data are used as input data for each round of training. These data features include vertex time series and fiber bundle features. The adjacency matrix includes the cortical vertex adjacency matrix corresponding to the T1-weighted data and the weighted adjacency matrix corresponding to the resting-state functional magnetic resonance imaging (fMRI) data. The connection weights of the cortical vertex adjacency matrix are mapped to the [0,1] distribution interval using softmax to obtain a normalized adjacency matrix, which also serves as a trainable variable for the label propagation algorithm in the fusion model. The fiber bundle features are input into the embedding layer for dimensionality reduction, resulting in dimensionality-reduced fiber bundle features. For example, this dimensionality-reduced fiber bundle feature matrix is ​​a fiber bundle feature matrix of vertices × 72.

[0149] The initial brain map fusion models include graph convolution models, label propagation models, and graph convolution models with masked inputs.

[0150] A graph convolution model is constructed, and the reduced fiber bundle features and the normalized adjacency matrix are input into the graph convolution model. The fiber bundle features are classified through graph convolution to obtain the first brain map prediction value.

[0151] A label propagation model is constructed by inputting the normalized adjacency matrix and the group graph labels into the label propagation model to obtain a set of second brain graph propagation values.

[0152] A prior mask and a graph convolutional model with masked input are constructed using vertex time series and group graph labels. The vertex time series, prior mask, and weighted adjacency matrix are fed into the graph convolutional model with masked input, and the predicted value of the third brain map is obtained through graph convolutional classification.

[0153] The predicted values ​​of the first and third brain maps are stacked together and passed through a linear layer to obtain the predicted value of the fourth brain map. Then, the cross-entropy between the predicted values ​​and the group map is calculated to obtain the first loss function. The propagated value of the second brain map is directly cross-entropy with the group map to obtain the second loss function. The second loss function is multiplied by a coefficient and added to the first loss function to obtain the final loss function. The final loss function is iterated until convergence, and the brain map fusion model training is completed.

[0154] The individual brain mapping module 53 is used to input the test data into the trained brain mapping fusion model to obtain the individual brain map.

[0155] The received brain data of the subject to be predicted is preprocessed to obtain test data including the adjacency matrix of the cortical vertices to be tested, the time series of the vertices to be tested, the weighted adjacency matrix to be tested, and the fiber tract features to be tested. This test data is then input into a trained brain mapping fusion model to obtain predicted values ​​for individual diffuse magnetic resonance imaging (DMRI) brain atlases, functional magnetic resonance imaging (fMRI) brain atlases, and fusion model predictions. An individual brain atlas is obtained based on the fusion model predictions. Additionally, individual DMRI brain atlases and functional MRI brain atlases can be obtained from the predicted values ​​of the individual DMRI brain atlases, serving as supplementary references for physicians. DMRI brain atlases can more accurately reflect brain structure, while fMRI brain atlases can more accurately reflect the status of active brain regions.

[0156] Each module in the aforementioned individual brain mapping device 50 based on a brain mapping fusion model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0157] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for drawing individual brain maps based on a brain map fusion model. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0158] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0159] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0160] Step S210: Receive at least two sets of subject data for training the brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0161] Step S220: Input the data features and adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation models to obtain the brain map prediction value output by the initial brain map fusion model. Train the initial brain map fusion model based on graph convolution and label propagation algorithms according to the brain map prediction value to obtain the trained brain map fusion model.

[0162] Step S230: Input the data to be tested into the trained brain map fusion model to obtain an individual brain map.

[0163] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0164] Step S210: Receive at least two sets of subject data for training the brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0165] Step S220: Input the data features and adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation models to obtain the brain map prediction value output by the initial brain map fusion model. Train the initial brain map fusion model based on graph convolution and label propagation algorithms according to the brain map prediction value to obtain the trained brain map fusion model.

[0166] Step S230: Input the data to be tested into the trained brain map fusion model to obtain an individual brain map.

[0167] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0168] Step S210: Receive at least two sets of subject data for training the brain map fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix.

[0169] Step S220: Input the data features and adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation models to obtain the brain map prediction value output by the initial brain map fusion model. Train the initial brain map fusion model based on graph convolution and label propagation algorithms according to the brain map prediction value to obtain the trained brain map fusion model.

[0170] Step S230: Input the data to be tested into the trained brain map fusion model to obtain an individual brain map.

[0171] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0172] 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 can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0173] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for drawing individual brain maps based on a brain map fusion model, characterized in that, The brain mapping method includes: The method includes receiving at least two sets of subject data for training a brain mapping fusion model, extracting data features from the subject data, and preprocessing the subject data to obtain an adjacency matrix. The method comprises: receiving at least two sets of subject data for training a brain mapping fusion model, each set of subject data including T1-weighted data, resting-state functional magnetic resonance (fMRI) data, and diffusion-weighted magnetic resonance (DMR) data; performing data feature extraction processing on the T1-weighted data and the resting-state fMRI data to obtain an adjacency matrix corresponding to the T1-weighted data and an adjacency matrix corresponding to the resting-state fMRI data; and extracting fiber bundle features from the DMR data. The data features and the adjacency matrix are input into an initial brain map fusion model based on a graph convolutional model and a label propagation model to obtain the brain map prediction value output by the initial brain map fusion model. The initial brain map fusion model is then trained based on the brain map prediction value to obtain a trained brain map fusion model. This includes: normalizing the adjacency matrix corresponding to the T1-weighted data to obtain a normalized adjacency matrix; reducing the dimensionality of the fiber bundle features to obtain dimensionality-reduced fiber bundle features; and constructing a first graph convolutional model, inputting the normalized adjacency matrix and the dimensionality-reduced fiber bundle features into the model. The first brain atlas prediction value is obtained by inputting the normalized adjacency matrix and the preset group atlas labels into the first graph convolutional model; the second brain atlas prediction value is obtained by inputting the preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data into the second graph convolutional model; the initial brain atlas fusion model is trained based on the first brain atlas prediction value, the second brain atlas prediction value, and the third brain atlas prediction value to obtain the trained brain atlas fusion model. The test data is input into the trained brain map fusion model to obtain an individual brain map.

2. The method for drawing individual brain maps based on a brain map fusion model according to claim 1, characterized in that, The step of performing data feature extraction processing on the T1-weighted data and the resting-state functional magnetic resonance imaging (fMRI) data to obtain the adjacency matrix corresponding to the T1-weighted data and the adjacency matrix corresponding to the resting-state fMRI data includes: The cortical surface structure is extracted from the T1 weighted data, the cortical surface structure is mapped to a standard template, and the cortical vertex adjacency matrix is ​​calculated based on the relationship between the triangular facet edges of the mapped cortical surface structure. Vertex time series are extracted from the resting-state functional magnetic resonance data. The Euclidean distance between any two vertices in the vertex time series is calculated. The dot product of the connection weights between the vertices and the Euclidean distances is calculated. The dot product is minimized iteratively to obtain a weighted adjacency matrix.

3. The method for drawing individual brain maps based on a brain map fusion model according to claim 1, characterized in that, The extraction of fiber bundle features from the diffusion-weighted magnetic resonance data includes: The initial fiber bundle features of the diffusion-weighted magnetic resonance data are extracted, the number of voxels of the initial fiber bundle features is traversed, and the initial fiber bundle features are filled according to the number of longest voxels to obtain the fiber bundle features.

4. The method for drawing individual brain maps based on a brain map fusion model according to claim 1, characterized in that, The construction of the second graph convolutional model involves inputting a preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data into the second graph convolutional model to obtain the third brain map prediction value, including: Construct a second graph convolutional model that includes two graph convolutional layers; The data features corresponding to the resting-state functional magnetic resonance data and the adjacency matrix corresponding to the resting-state functional magnetic resonance data are input into the second graph convolution model to obtain the output result of the second graph convolution model; The output is multiplied by a preset prior mask to obtain the predicted value of the third brain map.

5. The method for drawing individual brain maps based on a brain map fusion model according to claim 1, characterized in that, The initial brain map fusion model is trained based on the first brain map prediction value, the second brain map prediction value, and the third brain map prediction value to obtain a trained brain map fusion model, including: The first brain map prediction value and the third brain map prediction value are stacked to obtain a stacked result. The stacked result is then processed into a linear layer to obtain a fourth brain map prediction value. A first loss function is obtained by calculating the cross-entropy between the predicted value of the fourth brain map and the preset group map; a second loss function is obtained by calculating the cross-entropy between the predicted value of the second brain map and the preset group map. The first loss function and the second loss function are weighted and calculated to obtain the final loss function; Iterate the final loss function until it minimizes convergence to obtain the trained brain map fusion model.

6. An individual brain mapping device based on a brain mapping fusion model, characterized in that, The individual brain mapping device includes: A feature extraction module is used to receive at least two sets of subject data for training a brain mapping fusion model, extract data features from the subject data, and preprocess the subject data to obtain an adjacency matrix. The module includes: receiving at least two sets of subject data for training a brain mapping fusion model, each set of subject data including T1-weighted data, resting-state functional magnetic resonance (fMRI) data, and diffusion-weighted magnetic resonance (DMR) data; performing data feature extraction processing on the T1-weighted data and the resting-state fMRI data to obtain an adjacency matrix corresponding to the T1-weighted data and an adjacency matrix corresponding to the resting-state fMRI data; and extracting fiber bundle features from the DMR data. The training model module is used to input the data features and the adjacency matrix into the constructed initial brain map fusion model based on graph convolution and label propagation, to obtain the brain map prediction value output by the initial brain map fusion model. Based on the brain map prediction value, the initial brain map fusion model based on graph convolution and label propagation is trained to obtain the trained brain map fusion model. This includes: normalizing the adjacency matrix corresponding to the T1-weighted data to obtain a normalized adjacency matrix; reducing the dimensionality of the fiber bundle features to obtain dimensionality-reduced fiber bundle features; and constructing a first graph convolution model, inputting the normalized adjacency matrix and the dimensionality-reduced fiber bundle features into the model. The bundle features are input into the first graph convolutional model to obtain the first brain atlas prediction value; a label propagation model is constructed, and the normalized adjacency matrix and the preset group atlas labels are input into the label propagation model to obtain the second brain atlas prediction value; a second graph convolutional model is constructed, and the preset prior mask, the data features corresponding to the resting-state functional magnetic resonance imaging (fMRI) data, and the adjacency matrix corresponding to the resting-state fMRI data are input into the second graph convolutional model to obtain the third brain atlas prediction value; the initial brain atlas fusion model is trained based on the first brain atlas prediction value, the second brain atlas prediction value, and the third brain atlas prediction value to obtain the trained brain atlas fusion model; The individual brain mapping module is used to input the test data into the trained brain mapping fusion model to obtain the individual brain map.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the individual brain mapping method based on the brain mapping fusion model as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the individual brain mapping method based on the brain mapping fusion model as described in any one of claims 1 to 6.

Images