A method for generating a white matter fiber orientation distribution of a brain and related apparatus

Abstract

A method for generating white matter fiber direction distribution of brain and related device, comprising: obtaining diffusion magnetic resonance dMRI image data, and preprocessing the diffusion magnetic resonance dMRI image data; building a 3D feature extraction model to perform 3D feature extraction on the preprocessed data; building a 0-8 order spherical harmonic function coefficient SH estimation network model with the 3D feature vector as input; training the spherical harmonic function coefficient SH estimation network model, and generating the fiber direction distribution using the trained spherical harmonic function coefficient estimation network model. The present application realizes the generation of equivalent multi-shell multi-tissue constraint spherical deconvolution (MSMT-CSD) white matter fiber direction based on the deep learning method using single-shell three-tissue constraint spherical deconvolution (SS3T-CSD), not only estimates the fiber direction inside the central voxel, but also utilizes the spatial correlation between adjacent voxels, so that the accuracy of non-invasive prediction of white matter fiber direction is further improved.

Description

Technical Field

[0001] This invention belongs to the field of magnetic resonance medical image processing, and relates to a method and related apparatus for generating the directional distribution of white matter fibers in the brain. Background Technology

[0002] Traditional magnetic resonance imaging (MRI) techniques fail to accurately reconstruct the detailed fiber structures of different parts of the brain, lacking sufficient detail. Furthermore, invasive methods can cause irreversible damage to subjects or research participants. Diffusion-modulated magnetic resonance imaging (dMRI), on the other hand, can macroscopically reconstruct the connections between brain fiber bundles and remains the only non-invasive method for mapping the entire brain's fiber bundles.

[0003] In the field of diffusion-weighted magnetic resonance imaging (DMRI), tractography refers to the process of inferring the streamline structure of fibers, which is the result of local alignment with DMRI measurements. Since its advent, tractography has become a powerful tool for non-invasively studying the mechanisms behind normal and abnormal brain function. The workflow of brain tractography mainly consists of the following steps: preprocessing, fiber orientation resolution, fiber tracking, and post-tracking processing. In the fiber orientation resolution step, the fiber orientation distribution (FOD) predicted by constrained spherical deconvolution (CSD) is considered one of the most suitable methods for high-angle-resolution diffusion imaging (HARDI) of multiple shells. However, multi-shell HARDI data is not widely used in clinical settings due to limitations imposed by scanning equipment, scanning protocols, and acquisition time. In contrast, single-shell low angular resolution diffusion imaging (LARDI) is more commonly used in clinical practice. However, LARDI may produce false positive connections and lose some connections due to its insufficient angular resolution, resulting in insufficient accuracy of FOD prediction based on LARDI and insufficient ability to distinguish between multifiber crossings, bends, and kissing. Therefore, how to convert FOD predicted by single-shell LARDI, which is readily available in clinical practice, to FOD predicted by multi-shell HARDI is a problem worthy of in-depth research. Summary of the Invention

[0004] The purpose of this invention is to provide a method and related apparatus for generating the directional distribution of white matter fibers in the brain, so as to solve the problems of short scanning time but lack of radial information in single-shell scanning schemes; and the problems of long scanning time in multi-shell schemes, which solve the problem of lack of radial information.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for generating the directional distribution of white matter fibers in the brain, comprising:

[0007] Acquire diffusion magnetic resonance (dMRI) image data and preprocess the dMRI image data;

[0008] Build a 3D feature extraction model to extract 3D features from the preprocessed data;

[0009] Using 3D feature vectors as input, a network model for estimating the coefficients of spherical harmonic functions from order 0 to 8 is constructed.

[0010] Train the spherical harmonic function coefficient SH estimation network model, and use the trained spherical harmonic function coefficient estimation network model to generate fiber orientation distribution.

[0011] Optionally, preprocessing can be performed on the diffusion magnetic resonance (dMRI) image data:

[0012] Download the diffusion-weighted magnetic resonance (dMRI) image data. The dMRI image data uses the public dataset Human Connectome Project (HCP) dataset, which has 3T high angular resolution dMRI, acquired at an isotropic resolution of 1.25mm. The diffusion weight consists of three spherical shells with b = 1000, 2000, and 3000 s / mm². Each spherical shell acquires images in 90 directions and also has 18 images with b = 0 s / mm², for a total of 288 DWI images.

[0013] Optionally, the 3D feature extraction model includes a volume attention mechanism, a surface attention mechanism, a residual connection layer, a ReLU activation function, and a feedforward neural network. The volume attention mechanism is used to extract fiber orientation features in neighboring voxels, and the surface attention mechanism is used to extract fiber orientation features near each face of a voxel. The residual connection layer is used to solve the gradient vanishing problem. The feedforward neural network is used to enhance the model's capabilities. The extracted 3D features are used as input to the next linear layer.

[0014] Optionally, the volume attention mechanism uses the spatial correlation between the target voxel and its surrounding voxels to estimate the fiber orientation information contained in neighboring voxels; the surface attention mechanism starts from different surfaces of the target voxel to more accurately estimate the fiber orientation information contained in the voxel itself and its neighboring voxels. After two attention mechanisms, a residual connection layer is used to prevent gradient vanishing. After using the ReLU activation function, a feedforward neural network composed of three fully connected layers is used to enhance the fitting ability. Then, a residual connection layer is added again to enhance the learning features and reduce training error by combining the input channel and the feature extraction channel for further processing.

[0015] Optionally, the 0th to 8th order spherical harmonic function coefficient estimation module takes the 3D feature vector in the linear layer as input. Each estimation layer has four linear layers with sizes of 1024, 512, 256, and 128, respectively. Each linear layer has a batch normalization layer and a ReLU activation layer.

[0016] Optionally, based on the characteristics of spherical harmonic functions, a spherical harmonic function coefficient estimation module for orders 0 to 8 was designed, which is divided into 5 modules to calculate the spherical harmonic function coefficients for orders 0 to 8 respectively. The 3D features of the central voxel extracted from the Transformer-based module are passed through the estimation layer to calculate the spherical harmonic function coefficients of the central voxel.

[0017] Optionally, a network model for estimating spherical harmonic function coefficients (SH) is trained: first, clinically usable dMRI data is simulated through subsampling, and then single-shell low-angular-resolution dMRI is obtained through single-shell sampling. FOD data generated by this dMRI is used as the input to the entire network. In the model, rich 3D features within the voxels of interest and surrounding voxels are extracted. The 3D features extracted by each 3D feature extraction module are passed through two linear layers and then enter the 0 to 8th order SH coefficient estimation module to generate SH coefficients of each order. The FOD generated by single-shell low-angular-resolution dMRI and the FOD of MSMT-CSD as "ground reality" are input into the network for training. After repeated verification, it is made to meet the requirements.

[0018] In a second aspect, the present invention provides a system for generating the directional distribution of white matter fibers in the brain, comprising:

[0019] The data acquisition module is used to acquire diffusion magnetic resonance (dMRI) image data and preprocess the dMRI image data.

[0020] The feature extraction module is used to build a 3D feature extraction model and perform 3D feature extraction on the preprocessed data.

[0021] The network model building module is used to build a network model for estimating the coefficients of spherical harmonic functions from 0 to 8th order using 3D feature vectors as input.

[0022] The training output module is used to train the spherical harmonic function coefficient SH estimation network model, and the trained spherical harmonic function coefficient estimation network model is used to generate fiber orientation distribution.

[0023] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a method for generating the directional distribution of white matter fibers in the brain.

[0024] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for generating the directional distribution of white matter fibers in the brain.

[0025] Compared with the prior art, the present invention has the following technical effects:

[0026] Compared to other methods, the method of this invention not only improves upon the spatial correlation between voxels, which is easily overlooked by other methods, but also achieves greater accuracy in judging special cases such as multi-fiber crossings, bends, and "kissing" (i.e., interlocking). Furthermore, compared to other methods, the method of this invention calculates the coefficients of spherical harmonic functions from 0 to 8th order separately, avoiding the influence of different value ranges and further improving the accuracy of predicted FOD. Attached Figure Description

[0027] Figure 1 This is the overall flowchart of the present invention.

[0028] Figure 2 This is a schematic diagram of the overall structure of the present invention.

[0029] Figure 3 This is a schematic diagram of the 3D feature extraction module of the present invention.

[0030] Figure 4 This is a schematic diagram of two attention mechanisms in the 3D feature extraction module of the present invention.

[0031] Figure 5 This is a schematic diagram of the 0-8th order spherical harmonic function (SH) coefficient estimation module of the present invention.

[0032] Figure 6 This is a schematic diagram illustrating the prediction of fiber orientation distribution according to the present invention. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0034] An embodiment of the method for generating the orientation distribution of brain white matter fibers according to the present invention is as follows: Figure 1 As shown, the overall framework of the model is as follows: Figure 2 As shown, the steps are as follows:

[0035] Data preprocessing:

[0036] Diffusion-weighted magnetic resonance (dMRI) image data were downloaded from the Human Connectome Project (HCP) dataset, a public dataset. This dataset contains 3T high angular resolution dMRI images acquired at an isotropic resolution of 1.25 mm. The diffusion weights consist of three shells with b = 1000, 2000, and 3000 s / mm², each acquiring 90 gradient directions and 18 images with b = 0 s / mm², for a total of 288 DWI images. Since single-shell data is widely used in clinical settings, we downsampled the first 20 gradient directions of the dataset to combine the b = 1000 s / mm² shell with one b = 0 s / mm² image into a complete dataset, resulting in 21 DWI images per subject. After acquiring the DWI images for each subject, MRtrix3 was used to generate the fiber orientation distribution (FOD). The specific approach is as follows: First, the Dhollander algorithm is used to calculate the white matter response function for each subject's data. Then, single-shell three-tissue constrained spherical deconvolution (SS3T-CSD) and multi-shell multi-tissue spherical deconvolution (MSMT-CSD) are used to generate clinical MR image FOD and "ground truth" FOD. The SS3T-CSD FOD serves as the input to the 3D feature extraction module and as the training data for the entire model; while the MSMT-CSD FOD serves as the ground truth for the entire model. The input signal for each direction in the gradient direction N can be represented as:

[0037]

[0038] in, Represents the standard polar angle, 0 ≤ θ ≤ π. Y i Let c represent the basis of the i-th spherical harmonic function. i The coefficients of the corresponding spherical harmonic function are represented by R, which represents the number of FOD spherical harmonic basis coefficients. In this dataset, R = 45.

[0039] Build a 3D feature extraction module:

[0040] The entire spherical harmonic function coefficient estimation network is as follows: Figure 2 As shown, the structure of the 3D feature extraction module is as follows: Figure 3As shown, the model consists of an improved Transformer, whose structure includes a volume attention mechanism, a surface attention mechanism, a residual connection layer, a modified linear unit (ReLU) activation function, and a feedforward neural network. The volume attention mechanism extracts fiber orientation features from neighboring voxels, while the surface attention mechanism extracts fiber orientation features near each face of a voxel. The residual connection layer addresses the gradient vanishing problem, and the feedforward neural network enhances the model's capabilities. The extracted 3D features serve as input to the channel adjustment layer. The entire module's data processing flow is as follows: the input SS3T-CSD FOD enters the 3D feature extraction module in the form of SH coefficients. After passing through the volume and surface attention mechanisms, three matrices are randomly generated: WQ, WK, and WV, representing the linear projections of queries, keys, and values, respectively. The FOD in each direction is multiplied by these three matrices simultaneously as an SH coefficient vector to obtain 'Queries', 'Keys', and 'Values'. The difference between the two attention mechanisms is that the Q, K, and V obtained by the volume attention mechanism strengthen the relationship between the FOD within the central voxel and the FOD of surrounding voxels, as shown below. Figure 4 As shown in a; the surface attention mechanism is similar, such as Figure 4 As shown in b, the FOD of the central voxel and the surrounding voxels is predicted from the dimensions of each face. Then, the spatial relationship features between voxels can be obtained through residual connection and ReLU activation function, which plays a crucial role in the accuracy of FOD prediction.

[0041] Construct a module for estimating the coefficients of spherical harmonic functions from order 0 to 8:

[0042] The module structure diagram is as follows: Figure 5 As shown, the feature vector output from the channel adjustment layer passes through four multilayer perceptrons to calculate the 0th, 2nd, 4th, 6th, and 8th order SH coefficients (since FOD is symmetric, the odd-order spherical harmonic coefficients are all 0, and only the even-order spherical harmonic coefficients are calculated). Each of the first three linear layers is followed by a batch normalization layer and a ReLU activation layer with sizes of 1024, 512, 256, and 128, respectively. The calculated spherical harmonic coefficients are appended to the feature vector output from the previous module to calculate the coefficients for each order. This approach avoids the influence of different value ranges, resulting in higher accuracy in predicted FOD.

[0043] Training the spherical harmonic function (SH) estimation network model: First, the Dhollander algorithm is used to calculate the white matter response function for each subject's data. Then, single-shell three-tissue constrained spherical deconvolution (SS3T-CSD) and multi-shell multi-tissue spherical deconvolution (MSMT-CSD) are used to generate clinical MR image FOD and "ground reality" FOD. The SS3T-CSD FOD is used as input to the 3D feature extraction module and as the training set data for the entire model. Training is performed using data from 30 subjects each time. The model settings in this embodiment are as follows:

[0044] (4.1) Initialization: The Xavier method is used to alleviate the problem of gradient vanishing or exploding during backpropagation, and the convergence speed of the entire network will be faster.

[0045] (4.2) Activation Function: The Modified Linear Unit (ReLU) is used as the activation function in the 3D feature extraction module and the 0-8 order spherical harmonic function coefficient estimation module. Its definition is:

[0046] f(x) = max(0,x) (2)

[0047] (4.3) Loss Function: Since this problem involves a regression task, the loss function is defined as follows:

[0048]

[0049] Where P is the output FOD of the entire model, Q is the FOD obtained through MSMT-CSD, i.e. "ground-truth", R represents the number of spherical harmonic basis coefficients of FOD, and R = 45 in this dataset, and i is the index of the spherical harmonic basis coefficients.

[0050] FOD estimation using the trained model

[0051] The trained FOD estimation model is used to estimate FOD for each subject's data on the test set and validation set one by one until all subject data are estimated, thus completing the FOD estimation test.

[0052] In another embodiment of the present invention, a system for generating the directional distribution of brain white matter fibers is provided, which can be used to implement the above-described method for generating the directional distribution of brain white matter fibers. Specifically, the system includes:

[0053] The data acquisition module is used to acquire diffusion magnetic resonance (dMRI) image data and preprocess the dMRI image data.

[0054] The feature extraction module is used to build a 3D feature extraction model and perform 3D feature extraction on the preprocessed data.

[0055] The network model building module is used to build a network model for estimating the coefficients of spherical harmonic functions from 0 to 8th order using 3D feature vectors as input.

[0056] The training output module is used to train the spherical harmonic function coefficient SH estimation network model, and the trained spherical harmonic function coefficient estimation network model is used to generate fiber orientation distribution.

[0057] The data preprocessing module downloads diffusion magnetic resonance (dMRI) image data, calculates the white matter response function for each subject's data using the Dhollander algorithm, and then calculates the FOD using single-shell three-tissue spherical deconvolution, which is used as input to the entire spherical harmonic function coefficient estimation network.

[0058] The 3D feature extraction module is used to extract the spatial relationship features between the central voxel and its surrounding voxels, making the estimated FOD more accurate.

[0059] The module for estimating coefficients of spherical harmonic functions from 0 to 8th order is used to estimate the coefficients of spherical harmonic functions from 0 to 8th order respectively, avoiding the influence caused by different value ranges;

[0060] The spherical harmonic function (SH) coefficient estimation network model training module can improve the accuracy of the FOD estimation by the entire network through a certain number of training rounds, and reduce the risk of overfitting.

[0061] And a spherical harmonic function (SH) estimation module, which uses the trained model to estimate FOD.

[0062] To verify the beneficial effects of the present invention, the inventors conducted a qualitative comparative experiment on the HCP public dataset using the method of the embodiments of the present invention, along with SS3T-CSD and MSMT-CSD. Figure 6 As shown in Table 1, the proposed model has demonstrated competitive performance compared to MSMT-CSD. In the quantitative experiments, the angular correlation coefficient (ACC) was used as the evaluation metric to assess how close the white matter voxels and mixed plasm voxels are to the "ground truth". From the average value, our method achieved a high accuracy of 0.86 in predicting the white matter (WM) region. However, there is still considerable room for improvement in the prediction of the region between white matter and cortical gray matter (CGM). This is because our design of attention mechanisms has mostly focused on the prediction of fiber direction within the white matter, and there is still a lack of accuracy for the region outside the white matter. We will focus on solving this problem in our future work.

[0063] Table 1 shows the evaluation results using the angle correlation coefficient (ACC).

[0064]

[0065] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0066] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a method for generating the directional distribution of white matter fibers in the brain.

[0067] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for generating the directional distribution of white matter fibers in the brain in the above embodiments.

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

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

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

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

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for generating the directional distribution of white matter fibers in the brain, characterized in that, include: Acquire diffusion magnetic resonance (dMRI) image data and preprocess the dMRI image data; Build a 3D feature extraction model to extract 3D features from the preprocessed data; Using 3D feature vectors as input, a network model for estimating the coefficients of spherical harmonic functions from order 0 to 8 is constructed. Train the spherical harmonic function coefficient SH estimation network model, and use the trained spherical harmonic function coefficient estimation network model to generate fiber orientation distribution; The 3D feature extraction model includes volume attention mechanism, surface attention mechanism, residual connection layer, ReLU activation function, and feedforward neural network; the volume attention mechanism is used to extract fiber orientation features in neighboring voxels, and the surface attention mechanism is used to extract fiber orientation features near each face of a voxel; Residual connection layers are used to address the vanishing gradient problem; feedforward neural networks are used to enhance the model's capabilities; extracted 3D features are used as input to the next linear layer. The 0th to 8th order spherical harmonic function coefficient estimation module takes the 3D feature vector in the linear layer as input. Each estimation layer has four linear layers with sizes of 1024, 512, 256 and 128 respectively. Each linear layer has a batch normalization layer and a ReLU activation layer. Training the spherical harmonic function coefficient SH estimation network model: First, clinically usable dMRI data is simulated through subsampling. Then, single-shell low-angular-resolution dMRI is obtained through single-shell sampling. The FOD data generated by this dMRI is used as the input of the entire network. In the model, rich 3D features within the voxels of interest and surrounding voxels are extracted. The 3D features extracted by each 3D feature extraction module are passed through two linear layers and then enter the 0th to 8th order spherical harmonic function coefficient estimation module to generate SH coefficients of each order. The FOD generated by single-shell low-angular-resolution dMRI and the FOD of MSMT-CSD as "ground reality" are input into the network for training. After repeated verification, it is made to meet the requirements.

2. The method for generating the directional distribution of brain white matter fibers according to claim 1, characterized in that, Preprocessing of diffusion magnetic resonance (dMRI) image data: Download the diffusion-weighted magnetic resonance (dMRI) image data. The dMRI image data uses the public dataset Human Connectome Project (HCP) dataset, which has 3T high angular resolution dMRI, acquired at an isotropic resolution of 1.25 mm. The diffusion weight consists of three spherical shells with b=1000, 2000, and 3000 s / mm2. Each spherical shell acquires images in 90 directions and also has 18 images with b=0 s / mm2, for a total of 288 DWI images.

3. The method for generating the directional distribution of brain white matter fibers according to claim 1, characterized in that, The volume attention mechanism uses the spatial correlation between the target voxel and its surrounding voxels to estimate the fiber orientation information contained in neighboring voxels; the surface attention mechanism starts from different surfaces of the target voxel to more accurately estimate the fiber orientation information contained in the voxel itself and its neighboring voxels. After two attention mechanisms, a residual connection layer is used to prevent gradient vanishing. After using the ReLU activation function, a feedforward neural network composed of three fully connected layers is used to enhance the fitting ability. Then, a residual connection layer is added again to enhance the learning features and reduce training error by combining the input channel and the feature extraction channel for further processing.

4. The method for generating the directional distribution of brain white matter fibers according to claim 1, characterized in that, Based on the characteristics of spherical harmonic functions, a module for estimating the coefficients of spherical harmonic functions from order 0 to 8 was designed. The module is divided into 5 sub-modules to calculate the coefficients of spherical harmonic functions from order 0 to 8. The 3D features of the central voxel extracted from the Transformer-based module are passed through the estimation layer to calculate the coefficients of the central voxel.

5. A system for generating the directional distribution of white matter fibers in the brain, characterized in that, include: The data acquisition module is used to acquire diffusion magnetic resonance (dMRI) image data and preprocess the dMRI image data. The feature extraction module is used to build a 3D feature extraction model and perform 3D feature extraction on the preprocessed data. The network model building module is used to build a network model for estimating the coefficients of spherical harmonic functions from 0 to 8th order using 3D feature vectors as input. The training output module is used to train the spherical harmonic function coefficient SH estimation network model, and the trained spherical harmonic function coefficient estimation network model is used to generate fiber orientation distribution. The 3D feature extraction model includes volume attention mechanism, surface attention mechanism, residual connection layer, ReLU activation function, and feedforward neural network; the volume attention mechanism is used to extract fiber orientation features in neighboring voxels, and the surface attention mechanism is used to extract fiber orientation features near each face of a voxel; Residual connection layers are used to address the vanishing gradient problem; feedforward neural networks are used to enhance the model's capabilities; extracted 3D features are used as input to the next linear layer. The 0th to 8th order spherical harmonic function coefficient estimation module takes the 3D feature vector in the linear layer as input. Each estimation layer has four linear layers with sizes of 1024, 512, 256 and 128 respectively. Each linear layer has a batch normalization layer and a ReLU activation layer. Training the spherical harmonic function coefficient SH estimation network model: First, clinically usable dMRI data is simulated through subsampling. Then, single-shell low-angular-resolution dMRI is obtained through single-shell sampling. The FOD data generated by this dMRI is used as the input of the entire network. In the model, rich 3D features within the voxels of interest and surrounding voxels are extracted. The 3D features extracted by each 3D feature extraction module are passed through two linear layers and then enter the 0th to 8th order spherical harmonic function coefficient estimation module to generate SH coefficients of each order. The FOD generated by single-shell low-angular-resolution dMRI and the FOD of MSMT-CSD as "ground reality" are input into the network for training. After repeated verification, it is made to meet the requirements.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a method for generating the directional distribution of white matter fibers in the brain as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of a method for generating the directional distribution of white matter fibers in the brain as described in any one of claims 1 to 4.

Images