Motor-related electroencephalographic signal source localization method and system based on MNE-gan

Abstract

Disclosed in the present invention are a motor-related electroencephalographic signal source localization method and system based on an MNE-GAN. The method comprises the following steps: synthesizing virtual multi-channel electroencephalographic data; constructing a brain electrical source data generation network based on minimum norm estimation (MNE); constructing a brain electrical source data discriminator; preprocessing the virtual electroencephalographic data; training an MNE-GAN model on the basis of the virtual electroencephalographic data; and using the trained model to perform source localization on real electroencephalographic data. In the present invention, an MNE-based generative adversarial network (GAN) is used to provide a new brain electrical source localization method. A minimum norm constraint is added to restrict generated brain electrical source data to satisfy physical prior knowledge, thereby facilitating the research into non-invasive neurophysiological mechanisms and the improvement of the electroencephalographic decoding accuracy.

Description

A method and system for locating motor-related EEG signal sources based on MNE-GAN Technical Field

[0001] This invention relates to the field of brain power source localization technology, specifically to a method and system for localizing motor-related electroencephalogram (EEG) signal sources based on MNE-GAN. Background Technology

[0002] Electroencephalogram (EEG) is a technique that uses scalp electrodes to capture the weak electrical activity of neuronal groups in the cerebral cortex. Since German scientist Hans Berger first recorded this signal on the human scalp, its millisecond-level temporal resolution has made it a key tool for tracking dynamic changes in brain function (including cognitive process analysis, emotional state recognition, and motor intention decoding) and for clinical diagnosis (such as epileptic focus localization and sleep stage analysis). However, due to the effects of skull conductivity attenuation and volumetric conductor effects, the spatial resolution of scalp EEG signals is significantly limited, making it difficult to achieve precise spatial localization of intracranial neural activity.

[0003] To address the low spatial resolution of EEG signals, brain source localization technology has gradually developed. This technology inverts the three-dimensional spatial distribution characteristics of intracranial neural electrical activity from scalp potential distribution by establishing a forward head conduction model and solving an inverse problem. However, solving this problem is essentially a mathematical "inverse problem," with multiple solutions and a high dependence on the accuracy of the head conduction model. Existing source localization methods have significant limitations. Traditional algorithms based on Minimum Norm Estimation (MNE) and Standardized Low-Resolution Electromagnetic Tomography (sLORETA), while improving the solution space by introducing physical prior constraints, are prone to systematic localization errors due to oversimplification of the head geometry model. Secondly, traditional algorithms lack robustness and are sensitive to measurement noise such as electromyographic artifacts and electrode impedance changes, easily leading to distorted source localization results. In addition, existing methods are mostly based on static neural activity assumptions, making it difficult to effectively capture the temporal-spatial evolution characteristics of dynamic neural processes.

[0004] Deep learning has demonstrated significant advantages in brain power source localization, overcoming the limitations of traditional methods that rely on simplified models. It automatically learns the complex relationship between EEG signals and neural activity through a data-driven approach. Its dynamic modeling capabilities can track transient neural activity in real time, and combined with noise-resistant design and multimodal fusion technology, it effectively improves localization reliability. The end-to-end architecture also significantly improves computational efficiency, enabling real-time imaging and driving applications in precision medicine. These technical characteristics make deep learning an important research direction for optimizing brain power source localization performance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for locating motion-related EEG signal sources based on MNE-GAN. It introduces minimum norm constraints into deep neural networks for EEG source localization, offering a new possibility for deep analysis of EEG signals.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] Firstly, a method for locating motion-related EEG signal sources based on MNE-GAN includes the following steps:

[0008] (1) Based on the Brodmann partition, virtual brain power data under motion state is generated. Based on the real head model, the lead matrix is ​​obtained by calculating the forward propagation process of brain electricity. The virtual brain power data is mapped to the scalp electrode position. After the mapping process, electromyography, electrocardiography and electrooculography noise are added to generate brain electricity signals that simulate the real acquisition environment.

[0009] (2) Preprocessing of EEG signals, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and extraction of effective EEG features;

[0010] (3) Construct a generative adversarial network (MNE-GAN) model based on minimum norm estimation. In the MNE-GAN model, the generator uses preprocessed virtual EEG data as the training set, learns the spatial-temporal features of EEG signals through multi-layer convolution operations, outputs high-precision source spatial average EEG signals according to Brodmann segmentation, and establishes a mapping model from EEG signals to brain source distribution. The discriminator uses the source spatial average EEG signals generated by the generator as fake samples and the EEG signals simulated in step (1) as real samples for adversarial training. The binary cross-entropy loss is used as the loss function, and the generator loss is calculated through the discrimination results. The minimum norm inverse solution is introduced into the generator loss.

[0011] (4) The MNE-GAN model is trained using a phased training strategy: first, the discriminator is trained using a noisy signal, then the generator and discriminator are trained alternately until the model converges, and finally the minimum norm inverse solution is introduced as a supervision term for joint optimization.

[0012] (5) After preprocessing the real EEG data according to the method in step (2), input it into the trained MNE-GAN model. The model outputs the signal of the corresponding area of ​​the cerebral cortex, and then locates the area based on the intensity of the output signal.

[0013] Secondly, a motion-related electroencephalogram (EEG) signal source localization system based on MNE-GAN includes:

[0014] The virtual brain power data generation module is used to generate virtual brain power data under motion state according to Brodmann partition. It calculates the lead matrix based on the forward propagation process of EEG using a real head model, maps the virtual brain power data to the scalp electrode position, and adds electromyography, electrocardiography and electrooculography noise after the mapping process to generate EEG signals that simulate the real acquisition environment.

[0015] The EEG signal preprocessing module is used to preprocess EEG signals, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and to extract effective EEG features;

[0016] The MNE-GAN model construction module is used to construct a generative adversarial network (MNE-GAN) model based on minimum norm estimation. In the MNE-GAN model, the generator uses preprocessed virtual EEG data as the training set, learns the spatial-temporal features of EEG signals through multi-layer convolution operations, and outputs high-precision source spatial average EEG signals according to Brodmann segmentation, establishing a mapping model from EEG signals to brain source distribution. The discriminator uses the source spatial average EEG signals generated by the generator as fake samples and the EEG signals simulated by the virtual brain source data generation module as real samples for adversarial training. It adopts binary cross-entropy loss as the loss function, calculates the generator loss through the discrimination results, and introduces the minimum norm inverse solution into the generator loss.

[0017] The model training module is used to train the MNE-GAN model using a phased training strategy: first, the discriminator is trained using a noisy signal, then the generator and discriminator are trained alternately until the model converges, and finally the minimum norm inverse solution is introduced as a supervision term for joint optimization.

[0018] The source localization module is used to preprocess real EEG data according to the method of the EEG signal preprocessing module and then input it into the trained MNE-GAN model. The model outputs signals of the corresponding regions of the cerebral cortex, and then localization is performed based on the output signal intensity.

[0019] Thirdly, the present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the motion-related electroencephalogram (EEG) signal source localization method based on MNE-GAN as described in the first aspect.

[0020] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for locating motor-related electroencephalogram (EEG) signal sources based on MNE-GAN as described in the first aspect. Beneficial effects:

[0021] (1) This invention uses the Generative Adversarial Network (GAN) method to construct an inverse EEG signal solution model. Compared with the traditional brain power source localization method, this method breaks through the limitation of the traditional method relying on simplified models, which is conducive to enhancing the model localization accuracy and stability, and provides a new approach for brain power source localization method.

[0022] (2) This invention, through an adversarial training mechanism, can identify and suppress the influence of noise, thereby improving the robustness of the inverse solution results. Combined with the minimum norm constraint, it can effectively reduce the interference of noise on the inverse solution while ensuring the accuracy of the solution.

[0023] (3) This invention emphasizes the importance of simulating the generation of EEG signals. It uses the brain activity mechanism combined with Brodmann partitioning to design a method for simulating the generation of EEG signals, which can more reasonably simulate real brain activity.

[0024] (4) The present invention uses the minimum norm inverse solution as a constraint, which limits the range of the inverse solution, avoids overfitting and enhances the interpretability of the model. Attached Figure Description

[0025] Figure 1 is a main flowchart of the method of the present invention;

[0026] Figure 2 is a schematic diagram of brain regions related to movement;

[0027] Figure 3 is a schematic diagram of constructing the generator model;

[0028] Figure 4 is a schematic diagram of constructing the discriminator model. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0030] This invention provides a method for locating motor-related electroencephalogram (EEG) signal sources based on MNE-GAN. Referring to Figure 1, the method includes the following steps:

[0031] (1) Generating virtual brain power data: Using the MNE-Python toolkit, virtual brain power data is generated based on the Brodmann partition. The lead matrix is ​​obtained by calculating the forward propagation process of EEG based on the real head model. The virtual brain power data is mapped to the scalp electrode position. After the mapping process, electromyography, electrocardiography and electrooculography noise are added to generate EEG signal R(x) that simulates the real acquisition environment. The brain power data and simulated EEG data are saved.

[0032] Specifically, step (1) includes:

[0033] (1.1) A schematic diagram of motor-related brain regions is shown in Figure 2. The highlighted parts in the figure are 26 motor-related brain regions selected according to the Brodmann partition, including the primary somatic motor cortex (G_precentral), the extra / premotor area (S_precentral), the supplementary motor area (G_and_S_paracentral), the cortical-cortical motor integration zone (G_and_S_subcentral), and other limbic motor areas. In the upper limb motor-related regions, the primary somatic motor cortex and the extra / premotor area are set as the main activation areas, the postcentral gyrus (G_postcentral), the supplementary motor area (G_and_S_paracentral), the central sulcus (S_central), and the cortical-cortical motor integration zone (G_and_S_subcentral) are set as the synergistic areas, and the remaining motor-related regions are set as background areas.

[0034] (2.2) The dipoles of the selected 26 motor-related brain regions were initialized with a blank signal of 5 s duration and 256 Hz sampling frequency. In the selected main activation region, a sinusoidal wave with a random frequency of 8-13 Hz and a random amplitude of 5-10 μV was used to simulate the μ rhythm from 0 to 1 s, and a sinusoidal wave with a random frequency of 13-30 Hz and a random amplitude of 10-20 μV was used to simulate the β rhythm. From 1 to 3 s, the amplitudes of the μ and β rhythms were gradually reduced to 30% of their original amplitudes. From 3 to 5 s, the μ rhythm was kept unchanged, and the amplitude of the β rhythm was gradually increased back to its original amplitude. In the synergistic region, 50% of the amplitude of the main activation region was used to simulate EEG activity during movement. In the background region, 0.1-2 μV random noise was used to simulate background noise and other low-frequency activities. The dipoles of different regions were assigned values ​​using the corresponding signals to obtain the overall simulation signal.

[0035] (2.3) Based on the forward BEM head model, the dielectric conductivity of the three-layer model is set (0.3 S / m for scalp, 0.006 S / m for skull, and 0.3 S / m for brain). The forward solution of the EEG signal in the motion-related region is calculated, and the EEG signal in the source space is mapped onto the electrodes. During the process, ECG, EMG, EOG noise and random noise are added to simulate the real EEG signal.

[0036] (2) The EEG signals generated in step (1) are preprocessed, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and effective EEG features are extracted.

[0037] Specifically, step (2) includes:

[0038] (2.1) Perform a 0.3Hz-45Hz bandpass filter on the data to remove outliers, and use the channel data near the outlier channel to repair the outlier channel data using linear interpolation.

[0039] (2.2) Independent component analysis (ICA) algorithm was used to remove electromyography, electrocardiography and electrooculography interference from EEG signals;

[0040] (2.3) The processed signal is segmented and baseline correction and common average reference are performed.

[0041] (3) Construct a convolutional neural network (CNN) based on minimum norm estimation (MNE) as a generator. Referring to Figure 3, use the virtual EEG data preprocessed in step (2) as the training set, learn the spatial-temporal features of EEG signals through multi-layer convolution operations, and output high-precision source average EEG signals according to Brodmann segmentation and block output to establish a mapping model from EEG signals to brain source distribution.

[0042] Specifically, step (3) includes:

[0043] (3.1) The generator uses a convolutional neural network (CNN) combined with a fully connected layer to extract EEG signal features, and then uses the preprocessed EEG signals X∈R from c electrode channels and time t. c×t As input, it is passed to the convolutional layer for feature extraction, and then passed to the pooling layer for dimensionality reduction. The input feature of the k-th layer is h. k-1 The convolution calculation for this layer can be expressed as: h k =f(Conv(h) k-1 ,ω k ,b k f(·) represents the ReLU activation function, ω is the weight parameter, b is the bias, and pooling is performed after every two convolution operations to reduce the overall computational cost. A total of five convolution operations are used to finally obtain the EEG encoded data s.

[0044] (3.2) Decoding the data s encoded by the convolutional layer using fully connected layers, calculated using two fully connected layers, can be represented as h m =f(ω) m *h m-1 +b m f(·) represents the ReLU activation function, ω is the weight parameter, b is the bias, and a linear activation function is used to obtain the EEG features.

[0045] (3.3) The obtained EEG features are divided into blocks for mapping. Based on the Brodmann partitioning, the motion-related regions are selected, and each block is mapped individually using a single-layer convolution. Assuming that each block has r signal sources, the average EEG signal A(x)∈R from a single source is obtained through mapping. 1*t The average signal of r signal sources in the corresponding source space is spliced ​​together to output the overall average EEG signal G(x).

[0046] In the context of this invention, the source spatial average EEG signal and the source average EEG signal can be used interchangeably, referring to the EEG signal obtained by partition averaging. The source average EEG signal in step (1) refers to the signal obtained by averaging the dipole signals in each partition. The signal in step (3) is the source average EEG signal directly generated by the generator. The EEG signal refers to the initial signal of each partition containing multiple dipole signals without averaging.

[0047] (4) Construct a discriminator model. Referring to Figure 4, the source average EEG signal generated in step (3) is labeled as a fake sample (label 0). The brain power signal generated in step (1) is averaged by dipoles according to the partition to obtain the source average EEG signal, which is labeled as a real sample (label 1). The two are input into the discriminator for adversarial training. The discriminator parameters are updated by binary cross-entropy loss, and the generator loss is calculated by the discrimination result. The traditional minimum norm solution (MNE) is introduced into the generator loss as a supervision signal to enhance the physical rationality of the source spatial distribution.

[0048] Specifically, step (4) includes:

[0049] (4.1) Assume that the signal input to the discriminator is S(x). After passing through three convolutional layers and two fully connected layers, the signal is output through the Sigmoid activation function to obtain the probability D(S(x)) that the input sample is a real sample.

[0050] (4.2) The loss function of the discriminator is calculated using binary cross-entropy. The loss for real samples is -log(D(R(x))), and the probability of a real sample being correctly identified, D(R(x)), should ideally be 1. The loss for generated samples is -log(1-D(G(x))), and the probability of a generated sample being correctly identified as a real sample, D(G(x)), should ideally be 0. The final loss of the discriminator is the average of the sum of the loss for generated samples and the loss for real samples.

[0051] (4.3) The adversarial loss can be calculated using the probability D(S(x)) that the input sample output by the discriminator is a real sample:

[0052] To avoid large fluctuations in the early stages of training, the generator's initial loss function is only adversarial loss. Once the generation results stabilize, a minimum norm constraint loss function (MNE) is added.

[0053] Where G(x) i ) is the generator output. For data x iThe corresponding minimum norm inverse solution was calculated using a Python toolkit to obtain the inverse solution MNE(x). Source signals from motion-related regions were selected based on Brodmann partitioning as the final result, and the minimum norm inverse solution reference was obtained by averaging the region signals. Adding a minimum norm constraint loss provides explicit guidance for the generated EEG data, further enhancing its reliability. The final loss function is obtained by weighting the adversarial loss and the minimum norm constraint loss: Loss Generator =Loss GAN +γ mne *Loss mne

[0054] γ mne These are the weighting coefficients for the minimum norm loss, used to balance the effects of various losses.

[0055] (5) The MNE-GAN model is trained using a phased training strategy: the discriminator and generator are trained alternately until the model reaches initial convergence, and then physical constraints and supervision terms are introduced for joint optimization to ensure model stability and localization accuracy.

[0056] Specifically, step (5) includes:

[0057] (5.1) Pre-train the GAN network, fix the generator parameters, use noise as fake samples, and use the average EEG signal of the virtual source as real samples to iteratively update the discriminator parameters.

[0058] (5.2) Use a pre-trained discriminator and a well-designed generator for adversarial training, iterating the parameters alternately until convergence;

[0059] (5.3) After the model initially converges, a loss is introduced. mne γ mne Conduct joint training until Loss Generator The descent converges.

[0060] (6) Perform preprocessing on real EEG data, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and extract effective EEG features;

[0061] (7) Input the real EEG data processed in step (6) into the trained MNE-GAN model for source localization. The model outputs the collected EEG signals as signals of the corresponding regions of the cerebral cortex, i.e., source signals, and then performs localization based on signal intensity. Furthermore, the source signals obtained by this model and the source signals obtained by the MNE method are used simultaneously for an action classification task to compare the classification accuracy of the two methods. The generation effect is evaluated using classification accuracy.

[0062] Based on the same technical concept as the method embodiment, another embodiment of the present invention provides a motion-related electroencephalogram (EEG) signal source localization system based on MNE-GAN, comprising:

[0063] The virtual brain power data generation module is used to generate virtual brain power data under motion state according to Brodmann partition. It calculates the lead matrix based on the forward propagation process of EEG using a real head model, maps the virtual brain power data to the scalp electrode position, and adds electromyography, electrocardiography and electrooculography noise after the mapping process to generate EEG signals that simulate the real acquisition environment.

[0064] The EEG signal preprocessing module is used to preprocess EEG signals, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and to extract effective EEG features;

[0065] The MNE-GAN model construction module is used to construct a generative adversarial network (MNE-GAN) model based on minimum norm estimation. In the MNE-GAN model, the generator uses preprocessed virtual EEG data as the training set, learns the spatial-temporal features of EEG signals through multi-layer convolution operations, and outputs high-precision source spatial average EEG signals according to Brodmann segmentation, establishing a mapping model from EEG signals to brain source distribution. The discriminator uses the source spatial average EEG signals generated by the generator as fake samples and the EEG signals simulated by the virtual brain source data generation module as real samples for adversarial training. It adopts binary cross-entropy loss as the loss function, calculates the generator loss through the discrimination results, and introduces the minimum norm inverse solution into the generator loss.

[0066] The model training module is used to train the MNE-GAN model using a phased training strategy: first, the discriminator is trained using a noisy signal, then the generator and discriminator are trained alternately until the model converges, and finally the minimum norm inverse solution is introduced as a supervision term for joint optimization.

[0067] The source localization module is used to preprocess real EEG data according to the method of the EEG signal preprocessing module and then input it into the trained MNE-GAN model. The model outputs signals of the corresponding regions of the cerebral cortex, and then localization is performed based on the output signal intensity.

[0068] It should be understood that the motion-related EEG signal source localization system based on MNE-GAN in this embodiment can implement all the technical solutions in the above method embodiments. The functions of each functional module can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can be referred to the relevant descriptions in the above embodiments, which will not be repeated here.

[0069] Another embodiment of the present invention provides a computer device, including: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs, when executed by the processors, implement the steps of the MNE-GAN-based motion-related electroencephalogram (EEG) signal source localization method as described above.

[0070] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for locating motion-related electroencephalogram (EEG) signal sources based on MNE-GAN as described above.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), computer devices, 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.

[0072] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that each step in the flowchart and combinations of steps in the flowchart 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, create means for implementing the functions specified in one or more steps of the flowchart.

[0073] 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 that implement the functions specified in one or more processes of a flowchart.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more processes in the flowchart.

Claims

1. A method for locating motor-related electroencephalogram (EEG) signal sources based on MNE-GAN, characterized in that, Includes the following steps: (1) Based on the Brodmann partition, virtual brain power data under motion state is generated. Based on the real head model, the lead matrix is ​​obtained by calculating the forward propagation process of brain electricity. The virtual brain power data is mapped to the scalp electrode position. After the mapping process, electromyography, electrocardiography and electrooculography noise are added to generate brain electricity signals that simulate the real acquisition environment. (2) Preprocessing of EEG signals, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and extraction of effective EEG features; (3) Construct a generative adversarial network (MNE-GAN) model based on minimum norm estimation. In the MNE-GAN model, the generator uses preprocessed virtual EEG data as the training set, learns the spatial-temporal features of EEG signals through multi-layer convolution operations, outputs high-precision source spatial average EEG signals according to Brodmann segmentation, and establishes a mapping model from EEG signals to brain source distribution. The discriminator uses the source spatial average EEG signals generated by the generator as fake samples and the EEG signals simulated in step (1) as real samples for adversarial training. The binary cross-entropy loss is used as the loss function, and the generator loss is calculated through the discrimination results. The minimum norm inverse solution is introduced into the generator loss. (4) The MNE-GAN model is trained using a phased training strategy: first, the discriminator is trained using a noisy signal, then the generator and discriminator are trained alternately until the model converges, and finally the minimum norm inverse solution is introduced as a supervision term for joint optimization. (5) After preprocessing the real EEG data according to the method in step (2), input it into the trained MNE-GAN model. The model outputs the signal of the corresponding area of ​​the cerebral cortex, and then locates the area based on the intensity of the output signal.

2. The method according to claim 1, characterized in that, Step (1) includes: Based on the Brodmann partition, motor-related regions were selected, brain region labels were set, and motor-related regions were divided into main activation regions, synergistic regions, and background regions. The source dipole is initialized. In the selected activation region, sine waves of different frequencies and amplitudes are used in different partitions according to the label to simulate brain activity at different time periods. In the background region, random noise is used to simulate noise and low-frequency activity. The source dipole amplitude is given by the simulated signal. Based on the forward model, the forward solution of the EEG signal in the motion-related region is calculated, the EEG signal in the source space is mapped onto the electrodes, and ECG, EMG, EEG noise and random noise are added to simulate the real EEG signal.

3. The method according to claim 1, characterized in that, Step (2) includes: The data is bandpass filtered from 0.3Hz to 45Hz to remove outliers, and the abnormal channel data is repaired by interpolation. Independent component analysis (ICA) algorithm was used to remove electromyographic, electrocardiographic, and electrooculographic interference from electroencephalogram (EEG) signals. The processed signal is segmented and baseline correction and common average reference are performed.

4. The method according to claim 3, characterized in that, The generator uses a convolutional neural network combined with fully connected layers to extract EEG signal features, including: The preprocessed EEG signals X∈R from c electrode channels at time t are used to... c×t As input, it is passed to the convolutional layer for feature extraction, and then passed to the pooling layer for dimensionality reduction. The input feature of the k-th layer is h. k-1 The convolution calculation for this layer is represented as: h k =f(Conv(h) k-1 ,ω k ,b k f(·) represents the ReLU activation function, ω is the weight parameter, b is the bias, and finally the EEG encoded data s is obtained; Decoding the encoded data s from the convolutional layer using a fully connected layer is denoted as h. m =f(ω) m *h m-1 +b m ), ultimately yielding EEG characteristics; Using block mapping, motion-related regions are selected based on Brodmann partitioning, and each block is mapped individually. When each block has r signal sources in the source space, the mapping yields a single-source average EEG signal A(x)∈R. 1*t The average signal of r signal sources in the corresponding source space is spliced ​​together to output the overall average EEG signal G(x).

5. The method according to claim 1, characterized in that, In the discriminator model, the signal input to the discriminator is denoted as S(x). After passing through the convolutional layer and the fully connected layer, the signal is output through the Sigmoid activation function to obtain the probability D(S(x)) that the input sample is a real sample. The discriminator's loss function is calculated using binary cross-entropy. The loss for real samples is -log(D(R(x))), and the loss for generated samples is -log(1-D(G(x))). The final loss of the discriminator is the average of the two terms. Where N is the number of samples.

6. The method according to claim 5, characterized in that, The generator's loss is calculated as follows: the adversarial loss is calculated based on the probability D(S(x)) that the input sample output by the discriminator is a real sample. The generator's initial loss function is only adversarial loss; once the generation results stabilize, a minimum norm constraint loss function (MNE) is added. Where G(x) i ) is the generator output. For data x i The corresponding minimum norm inverse solution is calculated to obtain the inverse solution MNE(x). Based on the Brodmann partitioning, the source signal of the motion-related region is selected as the final result, and the minimum norm inverse solution reference is obtained by averaging the region signals. The final generator loss is obtained by weighting the adversarial loss and the minimum norm constraint loss: Loss Generator =Loss GAN +γ mne *Loss mne Where γ mne These are the weighting coefficients for the minimum norm loss, used to balance the effects of various losses.

7. The method according to claim 6, characterized in that, Step (5) includes: The model is pre-trained, the generator parameters are fixed, noise is used as fake samples, and the average EEG signal from the virtual source is used as real samples to iteratively update the discriminator parameters. Adversarial training is performed using a pre-trained discriminator and a well-designed generator, iterating the parameters alternately until convergence. Loss is introduced after the model initially converges. mne γ mne Joint training until Loss Generator The descent converges.

8. A motion-related electroencephalogram (EEG) signal source localization system based on MNE-GAN, characterized in that, include: The virtual brain power data generation module is used to generate virtual brain power data under motion state according to Brodmann partition. It calculates the lead matrix based on the forward propagation process of EEG using a real head model, maps the virtual brain power data to the scalp electrode position, and adds electromyography, electrocardiography and electrooculography noise after the mapping process to generate EEG signals that simulate the real acquisition environment. The EEG signal preprocessing module is used to preprocess EEG signals, including rereference, bandpass filtering, downsampling, artifact removal and slice segmentation, and to extract effective EEG features; The MNE-GAN model construction module is used to construct a generative adversarial network (MNE-GAN) model based on minimum norm estimation. In the MNE-GAN model, the generator uses preprocessed virtual EEG data as the training set, learns the spatial-temporal features of EEG signals through multi-layer convolution operations, and outputs high-precision source spatial average EEG signals according to Brodmann segmentation, establishing a mapping model from EEG signals to brain source distribution. The discriminator uses the source spatial average EEG signals generated by the generator as fake samples and the EEG signals simulated by the virtual brain source data generation module as real samples for adversarial training. It adopts binary cross-entropy loss as the loss function, calculates the generator loss through the discrimination results, and introduces the minimum norm inverse solution into the generator loss. The model training module is used to train the MNE-GAN model using a phased training strategy: first, the discriminator is trained using a noisy signal, then the generator and discriminator are trained alternately until the model converges, and finally the minimum norm inverse solution is introduced as a supervision term for joint optimization. The source localization module is used to preprocess real EEG data according to the method of the EEG signal preprocessing module and then input it into the trained MNE-GAN model. The model outputs signals of the corresponding regions of the cerebral cortex, and then localization is performed based on the output signal intensity.

9. A computer device, comprising: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the method for locating motor-related EEG signal sources based on MNE-GAN as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the method for locating motor-related electroencephalogram (EEG) signal sources based on MNE-GAN as described in any one of claims 1-7.

Images