An adaptive threshold depth unfolding brain-source imaging method and system

Abstract

This invention discloses an adaptive threshold-based depth-unfolding brain-source imaging method and system, relating to the field of brain-source imaging technology. The method includes: acquiring a training EEG signal observation matrix and a real source signal matrix; inputting the training EEG signal observation matrix into an initial iterative shrinking threshold depth-unfolding network, outputting a training reconstructed source signal matrix; determining a loss function value based on the real source signal matrix and the training reconstructed source signal matrix, and determining a target iterative shrinking threshold depth-unfolding network by minimizing the loss function value; acquiring a measured EEG signal observation matrix, solving for the measured EEG signal observation matrix using the target iterative shrinking threshold depth-unfolding network, and outputting a target reconstructed source signal matrix. Unfolding the traditional ISTA algorithm into a trainable deep neural network helps improve the reliability of brain-source imaging.

Description

Technical Field

[0001] This invention relates to the field of brain imaging technology, and more particularly to an adaptive threshold depth unfolding brain imaging method and system. Background Technology

[0002] Brain-derived imaging (BDI) technology, which visualizes brain electrical activity, is of great significance for cognitive neuroscience research, disease diagnosis and treatment, and the development of brain-computer interface systems. It includes invasive and non-invasive methods. Invasive BDI involves recording brain electrical activity by implanting electrodes or optical fibers. Because this requires implanting electrodes or fibers in the skull or under the skin, this invasive procedure can cause adverse reactions such as infection and bleeding. Furthermore, implants typically only detect the activity of neurons relatively close to them, failing to provide a comprehensive understanding of the activity in the entire brain region. The measurement results are also easily affected by the location and number of implants. Therefore, non-invasive BDI based on electroencephalography (EEG) and magnetoencephalography (MEG) to detect human brain function at millisecond temporal resolution is widely used.

[0003] SIFNet, a deep learning neural network for source imaging, is a non-invasive brain source imaging technique based on EEG or MEG signals for feature extraction and source reconstruction. Figure 1 As shown, the SIFNet network primarily uses a residual structure, cascading five residual blocks after a convolutional block, followed by a pooling layer and a fully connected layer. Its output size is the same as the number of source regions. Each residual block contains two Conv layers and undergoes batch normalization to eliminate covariance bias during training. The original input of each residual block is added back to the output and fed to the next module through a rectified linear unit (ReLU) activation function, transforming the brain source imaging problem into a classification problem. While the SIFNet network improves the accuracy of source reconstruction to some extent through deep learning, its reliance on convolutional operations can easily lead to the loss of signal details in certain situations. Its dependence on simulation data and a large amount of labeled data, coupled with its typically fixed regularization parameters, may limit the network's generalization ability. The network's training and feature learning processes are often black-box operations, leading to low confidence in the source reconstruction results. Furthermore, its high computational complexity reduces the overall reliability of brain source imaging. Summary of the Invention

[0004] This invention provides an adaptive threshold depth unfolding brain source imaging method and system, which improves the technical problem of low reliability of existing deep learning neural network source imaging methods for brain source imaging.

[0005] The first aspect of this invention provides an adaptive threshold depth-unfolding brain-source imaging method, comprising:

[0006] Obtain the training EEG signal observation matrix and the real source signal matrix;

[0007] The training EEG signal observation matrix is ​​used as input to the initial iterative shrinking threshold depth expansion network, and the training reconstructed source signal matrix is ​​output.

[0008] The loss function value is determined based on the real source signal matrix and the trained reconstructed source signal matrix, and the target iterative shrinkage threshold depth of the network is determined by minimizing the loss function value.

[0009] The measured EEG signal observation matrix is ​​collected, and the measured EEG signal observation matrix is ​​solved by a target iterative shrinking threshold depth expansion network to output the target reconstructed source signal matrix.

[0010] Optionally, obtaining the training EEG signal observation matrix and the real source signal matrix includes:

[0011] Acquire signals from multiple real sources;

[0012] Simulations were performed using the Brainstorm toolbox based on the real source signals, and the real source signal matrix and training lead field matrix were output.

[0013] The training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

[0014] Optionally, the step of using the trained EEG signal observation matrix as input to the initial iterative shrinking threshold depth unfolding network and outputting the trained reconstructed source signal matrix includes:

[0015] Construct a signal loss function for the EEG signal observation matrix;

[0016] Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are iteratively performed on the signal loss function to output the training reconstructed source signal matrix.

[0017] The signal loss function includes:

[0018] ;

[0019] In the formula, Let be the signal loss function. The square of the Frobenius norm. For the lead field matrix, The source signal matrix, This is the EEG signal observation matrix. For the first One source signal, The number of source signals, For regularization parameters, The first element of the source signal matrix One source signal, This represents the number of iterations.

[0020] The gradient descent operation includes:

[0021] ;

[0022] In the formula, For gradient, This is the Lipschitz constant. For the lead field matrix Transpose of;

[0023] The soft threshold calculation includes:

[0024] ;

[0025] in, , ;

[0026] In the formula, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, It is a symbolic function.

[0027] Optionally, the step of determining the loss function value based on the real source signal matrix and the trained reconstructed source signal matrix, and determining the target iterative shrinkage threshold depth of the unfolded network by minimizing the loss function value, includes:

[0028] The loss function value is calculated using the real source signal matrix and the trained reconstructed source signal matrix;

[0029] When the loss function value does not meet the early stopping condition, the model parameters are updated by using the Adam optimizer with L2 regularization to minimize the loss function value, and the hyperparameters are dynamically adjusted through cross-validation to determine the intermediate iteration shrinking threshold depth of the unfolded network.

[0030] The intermediate iterative shrinkage threshold depth unfolding network is trained using the training EEG signal observation matrix until the loss function value meets the early stopping condition, and the target iterative shrinkage threshold depth unfolding network is determined.

[0031] A second aspect of the present invention provides an adaptive threshold depth unfolding brain-source imaging system, comprising:

[0032] The training data acquisition module is used to acquire the training EEG signal observation matrix and the real source signal matrix;

[0033] The network training module is used to expand the network by taking the training EEG signal observation matrix as input and the initial iterative shrinking threshold depth, and output the training reconstructed source signal matrix.

[0034] The network iteration module is used to determine the loss function value based on the real source signal matrix and the trained reconstructed source signal matrix, and to determine the target iterative shrinkage threshold depth of the network by minimizing the loss function value;

[0035] The source signal reconstruction module is used to acquire the measured EEG signal observation matrix, solve the measured EEG signal observation matrix through a target iterative shrinking threshold depth expansion network, and output the target reconstructed source signal matrix.

[0036] Optionally, the training data acquisition module is specifically used for:

[0037] Acquire signals from multiple real sources;

[0038] Simulations were performed using the Brainstorm toolbox based on the real source signals, and the real source signal matrix and training lead field matrix were output.

[0039] The training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

[0040] Optionally, the network training module is specifically used for:

[0041] Construct a signal loss function for the EEG signal observation matrix;

[0042] Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are iteratively performed on the signal loss function to output the training reconstructed source signal matrix.

[0043] The signal loss function includes:

[0044] ;

[0045] In the formula, Let be the signal loss function. The square of the Frobenius norm. For the lead field matrix, The source signal matrix, This is the EEG signal observation matrix. For the first One source signal, The number of source signals, For regularization parameters, The first element of the source signal matrix One source signal, This represents the number of iterations.

[0046] The gradient descent operation includes:

[0047] ;

[0048] In the formula, For gradient, This is the Lipschitz constant. For the lead field matrix Transpose of;

[0049] The soft threshold calculation includes:

[0050] ;

[0051] in, , ;

[0052] In the formula, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, It is a symbolic function.

[0053] A computer device provided in a third aspect of the present invention includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any of the preceding claims.

[0054] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the adaptive threshold depth unfolding brain-source imaging method as described in any of the preceding claims.

[0055] The fifth aspect of the present invention provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the adaptive threshold depth unfolding brain-source imaging method as described in any of the preceding claims.

[0056] As can be seen from the above technical solutions, the present invention has the following advantages:

[0057] The above-described solution of the present invention provides an adaptive threshold depth-expanded brain-source imaging method, comprising: acquiring a training EEG signal observation matrix and a real source signal matrix; inputting the training EEG signal observation matrix into an initial iterative shrinking threshold depth-expanding network, and outputting a training reconstructed source signal matrix; determining a loss function value based on the real source signal matrix and the training reconstructed source signal matrix, and determining a target iterative shrinking threshold depth-expanding network by minimizing the loss function value; acquiring a measured EEG signal observation matrix, solving the measured EEG signal observation matrix through the target iterative shrinking threshold depth-expanding network, and outputting a target reconstructed source signal matrix. This method expands the traditional ISTA algorithm into a trainable deep neural network. Through iterative training, the regularization parameters for signal reconstruction can be adaptively adjusted, which is beneficial for improving the accuracy of brain-source imaging. It can explicitly demonstrate the optimization process of each iteration, making it interpretable. Each optimization update can efficiently process multiple iterations in parallel, improving computational efficiency. These advantages contribute to improving the reliability of brain-source imaging. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 Here is a diagram of the SIFNet network structure;

[0060] Figure 2 A flowchart illustrating the steps of an adaptive threshold depth unfolding brain-source imaging method provided in this embodiment of the invention;

[0061] Figure 3 A simplified schematic diagram of a depth unfolded network provided in an embodiment of the present invention;

[0062] Figure 4 This is a structural block diagram of an adaptive threshold depth unfolding brain-source imaging system provided in an embodiment of the present invention. Detailed Implementation

[0063] This invention provides an adaptive threshold depth unfolding brain source imaging method and system to improve the technical problem of low reliability in existing deep learning neural network source imaging methods for brain source imaging.

[0064] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0065] Please see Figure 2 , Figure 2 A flowchart illustrating the steps of an adaptive threshold depth unfolding brain-source imaging method provided in an embodiment of the present invention.

[0066] This invention provides an adaptive threshold depth unfolding brain-source imaging method, comprising:

[0067] Step 101: Obtain the training EEG signal observation matrix and the real source signal matrix.

[0068] Step 101 includes the following sub-steps:

[0069] Acquire signals from multiple real sources;

[0070] Simulations were performed using the Brainstorm toolbox based on various real source signals, and the real source signal matrix and training lead field matrix were output.

[0071] A training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

[0072] It should be noted that a training set is constructed using a training EEG signal observation matrix and a real source signal matrix to facilitate subsequent network training. In one implementation, this can be directly obtained from known data. In another implementation, multiple real source signals can be identified first, and the real source signal matrix and training lead field matrix can be determined based on each real source signal using the Brainstorm toolbox. The brain model is loaded using the Brainstorm toolbox, such as using the ICBM 152 default MRI template. After downsampling the ICBM 152 default MRI template to 6004 voxels, modeling is performed to determine the brain model. The brain model has multiple potential dipoles. One or more dipoles are randomly selected as seed dipoles in the cortical grid of the brain model, and adjacent units are iteratively absorbed into the patch based on various seed dipoles to expand until the required activation area is met, thus obtaining the source space. The direction of the neural current of each dipole in the source space is set to be perpendicular to the cortical surface, and real source signals are assigned to each dipole. The real source signal matrix is ​​then formed based on these real source signals. This is based on a 64-channel Neuroscan. The sensor configuration in the Quik-cap system uses the OpenMEEG software package embedded in Brainstorm to calculate and determine the lead field matrix as the training lead field matrix. Two reference electrodes are removed, and a three-layer head surface is obtained using the boundary element method (BEM). The specific process can be found in existing techniques. Based on the obtained training lead field matrix, the observation model is then used... A linear transformation is performed on the real source signal matrix and noise is added, where, This is the EEG signal observation matrix. For the lead field matrix, The source signal matrix, Noise is used to generate a training EEG signal observation matrix.

[0073] Step 102: Input the training EEG signal observation matrix into the initial iterative shrinking threshold depth expansion network and output the training reconstructed source signal matrix.

[0074] Step 102 includes the following sub-steps:

[0075] Construct a signal loss function for the EEG signal observation matrix;

[0076] Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are performed iteratively on the signal loss function to output the training reconstructed source signal matrix.

[0077] It should be noted that deep unfolded networks are a method that combines traditional optimization algorithms with deep learning. A simplified diagram is shown below. Figure 3As shown, it expands each step of the iterative algorithm into a layer of a deep network, forming an end-to-end trainable network structure. This method aims to inherit the interpretability of traditional optimization algorithms while leveraging the powerful modeling capabilities of deep learning. In this embodiment, the initial iterative shrinking threshold deep unfolded network is constructed by expanding the iterative process of ISTA (Iterative Shrinking Thresholding Algorithm) into a layer structure of a deep neural network.

[0078] 1) Input layer: Input EEG signal observation matrix ;

[0079] 2) Hidden Layer: To simultaneously consider data fidelity and the sparsity of the source signal, a signal loss function is defined:

[0080] ;

[0081] In the formula, Let be the signal loss function. The square of the Frobenius norm. The regularization parameter controls the degree of sparsity. For the first One source signal, The number of source signals, The first element of the source signal matrix One source signal, This represents the number of iterations.

[0082] The above optimization problem is solved by iteratively shrinking the threshold. The iterative process of ISTA is expanded into a layer structure of a deep neural network. Each layer corresponds to one iterative update. Each iterative update performs two operations: linear transformation and nonlinear transformation. Gradient descent is performed in the linear transformation, and soft thresholding is performed in the nonlinear transformation to achieve sparsity constraints.

[0083] Gradient descent operations include:

[0084] ;

[0085] In the formula, For the number of iterations, For gradient, The source signal matrix, This is the Lipschitz constant. For the lead field matrix transpose, This is the EEG signal observation matrix;

[0086] Soft thresholding includes:

[0087] ;

[0088] in, , ;

[0089] In the formula, For regularization parameters, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, For sign functions. The soft threshold operator is achieved by... Less than the threshold The component is set to 0, while the remaining components are reduced by amplitude. Implement sparsity constraints;

[0090] 3) Output layer: After multiple iterations and updates, if the signal iteration termination condition is met, the reconstructed source signal matrix is ​​output based on the optimal solution of the current signal loss function. .

[0091] Step 103: Determine the loss function value based on the real source signal matrix and the training reconstructed source signal matrix, and determine the target iterative shrinkage threshold depth of the network by minimizing the loss function value.

[0092] Step 103 includes the following sub-steps:

[0093] The loss function value is calculated using the real source signal matrix and the trained reconstructed source signal matrix;

[0094] When the loss function value does not meet the early stopping condition, the model parameters are updated by using the Adam optimizer with L2 regularization to minimize the loss function value, and the hyperparameters are dynamically adjusted through cross-validation to determine the intermediate iteration shrinkage threshold depth of the network.

[0095] The intermediate iterative shrinkage threshold depth unfolding network was trained using the training EEG signal observation matrix until the loss function value met the early stopping condition, thus determining the target iterative shrinkage threshold depth unfolding network.

[0096] It should be noted that the training data is used to train the network expansion model at the initial iteration shrinking threshold depth. The mean squared error loss function is used to measure the difference between the trained reconstructed source signal matrix and the true source signal matrix.

[0097] ;

[0098] In the formula, The value of the loss function. For the first training samples, The number of training samples. For the first The true source signal matrix of each training sample For the first The training reconstructed source signal matrix of each training sample;

[0099] An early stopping strategy is adopted during training. When the loss function value does not meet the early stopping condition, the Adam optimizer with adaptive matrix estimation is used to update the model parameters with the goal of minimizing the loss function value. L2 regularization is introduced during the model parameter update process to prevent overfitting. At the same time, hyperparameters such as the learning rate are dynamically adjusted through cross-validation. After obtaining the optimized intermediate iteration shrinking threshold depth unfolded network, the training data is used again for training until the loss function value meets the early stopping condition. Then, the target iteration shrinking threshold depth unfolded network is determined so that it can effectively reconstruct the source signal.

[0100] During network training, the model parameters, including weights and threshold parameters, are iteratively updated. The threshold parameters of each layer are learned and adjusted based on backpropagation of the loss function values. This allows the network to dynamically adjust the sparsity regularization parameters according to the characteristics of different data (such as noise level and signal strength). By introducing adaptive thresholding regularization, the network can better adapt to different sparsity, noise levels, and the non-uniformity of brain source signals. For strong signal components, the adaptive threshold can be lowered to reduce the suppression of the real signal. For weak signal or noise components, the adaptive threshold can be raised to better suppress noise. This makes the network more noise-adaptive and data-robust, and can provide higher source reconstruction accuracy under different data quality and noise environments. It is particularly suitable for EEG / MEG data with high noise in clinical settings.

[0101] Step 104: Collect the measured EEG signal observation matrix, solve the measured EEG signal observation matrix through the target iterative shrinking threshold depth expansion network, and output the target reconstructed source signal matrix.

[0102] It should be noted that the collected measured EEG signal observation matrix is ​​input into the trained target iterative shrinking threshold depth unfolding network. After forward propagation, the target reconstruction source signal matrix is ​​output. Through the nonlinear transformation of the deep network, the step effect in traditional methods is reduced, and more important detailed information is preserved.

[0103] In this embodiment of the invention, the traditional ISTA algorithm is unfolded into a trainable deep neural network, which shows great promise for improving the reliability of brain source imaging, and is particularly suitable for brain source imaging tasks that require high precision and real-time processing: 1. It can explicitly show the optimization process of each iteration, improving the transparency of the source signal reconstruction process, allowing users to understand how each layer adjusts the signal and why the network makes specific decisions; 2. It can efficiently process multiple iterations in parallel during each optimization update, avoiding the computational bottleneck of step-by-step iteration in traditional ISTA, significantly improving computational efficiency, especially when processing large-scale data; 3. This method combines the advantages of model-based and data-driven methods, and can optimize the reconstruction process by adaptively adjusting the regularization parameters of signal reconstruction through network iterative training while maintaining physical model constraints, thereby significantly improving the accuracy of brain source imaging while avoiding the invasive operational risks of invasive brain source imaging.

[0104] Please see Figure 4 , Figure 4 This is a structural block diagram of an adaptive threshold depth unfolding brain-source imaging system provided in an embodiment of the present invention.

[0105] This invention provides an adaptive threshold depth unfolding brain-derived imaging system, comprising:

[0106] The training data acquisition module 401 is used to acquire the training EEG signal observation matrix and the real source signal matrix;

[0107] Network training module 402 is used to expand the network by taking the training EEG signal observation matrix as input to the initial iterative shrinking threshold depth, and output the training reconstructed source signal matrix.

[0108] The network iteration module 403 is used to determine the loss function value based on the real source signal matrix and the trained reconstructed source signal matrix, and to determine the target iterative shrinkage threshold depth of the network by minimizing the loss function value.

[0109] The source signal reconstruction module 404 is used to acquire the measured EEG signal observation matrix, solve the measured EEG signal observation matrix through a target iterative shrinking threshold depth expansion network, and output the target reconstructed source signal matrix.

[0110] Optionally, the training data acquisition module 401 is specifically used for:

[0111] Acquire signals from multiple real sources;

[0112] Simulations were performed using the Brainstorm toolbox based on various real source signals, and the real source signal matrix and training lead field matrix were output.

[0113] A training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

[0114] Optionally, the network training module 402 is specifically used for:

[0115] Construct a signal loss function for the EEG signal observation matrix;

[0116] Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are performed iteratively on the signal loss function to output the training reconstructed source signal matrix;

[0117] Signal loss functions include:

[0118] ;

[0119] In the formula, Let be the signal loss function. The square of the Frobenius norm. For the lead field matrix, The source signal matrix, This is the EEG signal observation matrix. For the first One source signal, The number of source signals, For regularization parameters, The first element of the source signal matrix One source signal, This represents the number of iterations.

[0120] Gradient descent operations include:

[0121] ;

[0122] In the formula, For gradient, This is the Lipschitz constant. For the lead field matrix Transpose of;

[0123] Soft thresholding includes:

[0124] ;

[0125] in, , ;

[0126] In the formula, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, It is a symbolic function.

[0127] Optionally, the network iteration module 403 is specifically used for:

[0128] The loss function value is calculated using the real source signal matrix and the trained reconstructed source signal matrix;

[0129] When the loss function value does not meet the early stopping condition, the model parameters are updated by using the Adam optimizer with L2 regularization to minimize the loss function value, and the hyperparameters are dynamically adjusted through cross-validation to determine the intermediate iteration shrinkage threshold depth of the network.

[0130] The intermediate iterative shrinkage threshold depth unfolding network was trained using the training EEG signal observation matrix until the loss function value met the early stopping condition, thus determining the target iterative shrinkage threshold depth unfolding network.

[0131] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any of the above embodiments.

[0132] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implement the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any of the above embodiments.

[0133] This invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any of the above embodiments.

[0134] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0135] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0139] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A depth-expanding brain-source imaging method with adaptive thresholding, characterized in that, include: Obtain the training EEG signal observation matrix and the real source signal matrix; The training EEG signal observation matrix is ​​used as input to the initial iterative shrinking threshold depth expansion network, and the output is the training reconstructed source signal matrix; The loss function value is determined based on the real source signal matrix and the trained reconstructed source signal matrix, and the target iterative shrinkage threshold depth of the network is determined by minimizing the loss function value. The measured EEG signal observation matrix is ​​collected, and the measured EEG signal observation matrix is ​​solved by a target iterative shrinking threshold depth expansion network to output the target reconstructed source signal matrix; The process of using the trained EEG signal observation matrix as input to the initial iterative shrinking threshold depth expansion network and outputting a trained reconstructed source signal matrix includes: Construct a signal loss function for the EEG signal observation matrix; Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are iteratively performed on the signal loss function to output the training reconstructed source signal matrix. The signal loss function includes: ； Where, Let be the signal loss function. The square of the Frobenius norm. For the lead field matrix, The source signal matrix, This is the EEG signal observation matrix. For the One source signal, The number of source signals, For regularization parameters, The first element of the source signal matrix One source signal, This represents the number of iterations. The gradient descent operation includes: ； Where, For gradient, This is the Lipschitz constant. For the lead field matrix transpose; The soft threshold calculation includes: ； in, , ; In the formula, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, It is a symbolic function.

2. The adaptive threshold depth unfolding brain-source imaging method according to claim 1, characterized in that, The acquisition of the training EEG signal observation matrix and the real source signal matrix includes: Acquire signals from multiple real sources; Simulations were performed using the Brainstorm toolbox based on the real source signals, and the real source signal matrix and training lead field matrix were output. The training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

3. The adaptive threshold depth unfolding brain-source imaging method according to claim 1, characterized in that, The step of determining the loss function value based on the real source signal matrix and the trained reconstructed source signal matrix, and determining the target iterative shrinking threshold depth of the network by minimizing the loss function value, includes: The loss function value is calculated using the real source signal matrix and the trained reconstructed source signal matrix; When the loss function value does not meet the early stopping condition, the model parameters are updated by using the Adam optimizer with L2 regularization to minimize the loss function value, and the hyperparameters are dynamically adjusted through cross-validation to determine the intermediate iteration shrinking threshold depth of the unfolded network. The intermediate iterative shrinkage threshold depth unfolding network is trained using the training EEG signal observation matrix until the loss function value meets the early stopping condition, and the target iterative shrinkage threshold depth unfolding network is determined.

4. A depth-expanding brain-derived imaging system with adaptive threshold, characterized in that, include: The training data acquisition module is used to acquire the training EEG signal observation matrix and the real source signal matrix; The network training module is used to expand the network by taking the training EEG signal observation matrix as input and the initial iterative shrinking threshold depth, and output the training reconstructed source signal matrix. The network iteration module is used to determine the loss function value based on the real source signal matrix and the trained reconstructed source signal matrix, and to determine the target iterative shrinkage threshold depth of the network by minimizing the loss function value; The source signal reconstruction module is used to acquire the measured EEG signal observation matrix, solve the measured EEG signal observation matrix through a target iterative shrinking threshold depth expansion network, and output the target reconstructed source signal matrix. The network training module is specifically used for: Construct a signal loss function for the EEG signal observation matrix; Based on the training EEG signal observation matrix, gradient descent and soft thresholding operations are iteratively performed on the signal loss function to output the training reconstructed source signal matrix. The signal loss function includes: ； Where, Let be the signal loss function. The square of the Frobenius norm. For the lead field matrix, The source signal matrix, This is the EEG signal observation matrix. For the One source signal, The number of source signals, For regularization parameters, The first element of the source signal matrix One source signal, This represents the number of iterations. The gradient descent operation includes: ； Where, For gradient, This is the Lipschitz constant. For the lead field matrix transpose; The soft threshold calculation includes: ； in, , ; In the formula, For the variables of the soft threshold operator, For threshold parameters, For soft thresholding operators, It is a symbolic function.

5. The adaptive threshold depth unfolding brain-source imaging system according to claim 4, characterized in that, The training data acquisition module is specifically used for: Acquire signals from multiple real sources; Simulations were performed using the Brainstorm toolbox based on the real source signals, and the real source signal matrix and training lead field matrix were output. The training EEG signal observation matrix is ​​generated by linearly transforming the real source signal matrix based on the training lead field matrix and adding noise.

6. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any one of claims 1-3.

7. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any one of claims 1-3.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the adaptive threshold depth unfolding brain-source imaging method as described in any one of claims 1-3.

Images