Specific neurofeedback system for improving anxiety based on multi-modal fusion

Abstract

The application provides a specific neurofeedback training system for improving anxiety based on multi-modal fusion, which forms specific electroencephalogram signals through the projection strategy of magnetoencephalogram signals to electroencephalogram, and is used for the regulation and improvement of anxiety emotion. The application belongs to the technical field of medical treatment, and comprises an electroencephalogram acquisition module, a real-time processing module and a visual feedback module which are connected with each other; a complete treatment closed loop from signal acquisition, signal processing, signal feedback and signal acquisition is realized. The application decodes and analyzes electroencephalogram signals in real time, extracts features, and uses a specific mapping model of multi-modal feature fusion to establish a specific electroencephalogram mapping signal reflecting the real-time activity of core brain regions related to emotion (such as amygdala), which has stronger spatial accuracy and symptom specificity; the specific mapping signal is applied to neurofeedback treatment, which can assist in formulating individualized and multi-course neuroregulation training, improving the anxiety emotion of users, and providing a non-invasive and convenient neuroregulation platform for users.

Description

Technical Field

[0001] This invention belongs to the field of medical technology and relates to a specific neurofeedback system for improving anxiety based on multimodal fusion. Specifically, it relates to a specific signal index fused from electroencephalography (EEG) and magnetoencephalography (MEG) for use in a neurofeedback therapy system to improve anxiety. Background Technology

[0002] Anxiety disorder is a state of restlessness caused by excessive worry about the safety of loved ones or oneself, one's future, or destiny. Also known as anxiety neurosis, it is characterized primarily by the experience of anxiety. Main symptoms include: tension and worry without a clear objective object, restlessness, etc. In severe cases, it can cause immense subjective distress and even affect the patient's ability to adapt and integrate into the real social environment.

[0003] Current treatments for anxieties typically involve medication and psychotherapy. Medication primarily uses benzodiazepines (also known as tranquilizers), antidepressants, and a combination of long- and short-acting medications. Because these medications have relatively short durations of action, they are unsuitable for long-term, high-dose use and can lead to dependence, often placing a significant psychological and financial burden on patients. Psychotherapy, on the other hand, involves clinicians using professional verbal and non-verbal communication to guide and help patients change behavioral habits, cognitive coping mechanisms, etc. This requires a high level of expertise from the physician and cannot directly address neurobiological changes in the brain.

[0004] Increasingly, anxiety therapies are employing biofeedback therapy combined with psychotherapy, or relaxation and meditation, to guide subjects in alleviating anxiety. This has proven highly effective and can significantly improve patients' ability to regulate their emotions. Neurofeedback is one type of biofeedback technique that uses visual, auditory, or other methods to represent a specific neural activity and provides real-time feedback to the subject. The subject then uses self-regulation of this representation to improve behaviors or symptoms associated with that neural activity. The core of neurofeedback technology lies in capturing and feeding back the subject's neural activity. This detection method typically includes electroencephalography (EEG), magnetoencephalography (MEG), and hemodynamic imaging techniques such as functional magnetic resonance imaging (fMRI) and functional near-infrared imaging (fNIRS). Most mainstream research utilizes EEG and fMRI for neurofeedback studies.

[0005] Traditional neurofeedback methods use single-channel frequency signals from EEG for feedback, commonly employing alpha or beta frequency signals from channels such as P3, P4, F3, and F4 to regulate depressive mood, anxiety, and cognitive performance. However, EEG has significant limitations in localization; its low spatial resolution makes assessing the source of neural electrical activity through cranial measurements an ill-posed inverse problem.

[0006] In contrast, fMRI signals, through image acquisition and processing, can precisely acquire blood oxygenation level dependent (BOLD) signals from deep brain regions, exhibiting greater specificity and providing a novel approach to exploring the relationship between structural neural plasticity changes and cognitive behavioral functions in deep brain regions. However, fMRI signals have poor temporal resolution (on the order of seconds), which significantly reduces the real-time performance of neurofeedback training using fMRI signals, thus greatly diminishing the effectiveness of specific neurofeedback using fMRI.

[0007] Therefore, there is an urgent need to develop an innovative neurofeedback technology that can accurately target the patient's emotional areas while maintaining real-time performance, thus creating a training platform for patients to regulate their emotions autonomously. Summary of the Invention

[0008] To address the above problems, this invention provides a specific neurofeedback system for improving anxiety based on multimodal fusion.

[0009] The technical solution of the present invention is: a specific neurofeedback system for improving anxiety based on multimodal fusion of magnetoencephalography (MEG) and electroencephalography (EEG), comprising an EEG acquisition module, a real-time processing module, and a visual feedback module;

[0010] The EEG acquisition module is connected to the real-time processing module, the real-time processing module is connected to the visual feedback module, and the visual feedback module is connected to the EEG acquisition module, forming a closed-loop system. This system can be used to assist in developing individualized neurofeedback modulation training to improve the user's anxiety.

[0011] The electroencephalogram (EEG) acquisition module acquires and transmits EEG signals through hardware devices; it includes an acquisition unit and a communication unit 1. The acquisition unit acquires and encapsulates the EEG signals and then sends them to the real-time processing module through the communication unit 1.

[0012] The real-time processing module performs real-time decoding analysis and feature extraction of EEG signals, and uses an EEG-MEG multimodal fusion model to construct real-time targeted specific signals of the activity of emotion-related core brain regions (such as the amygdala) based on real-time EEG signals.

[0013] It includes a communication unit 2, a data preprocessing module, a feature extraction unit, and a multimodal specific signal construction unit. After acquiring EEG data from the EEG acquisition module, the communication unit 2 preprocesses the EEG signals through the data preprocessing module, and then further decodes them through the feature extraction unit. Through the multimodal specific signal construction unit, the decoded signals are projected onto the individual EEG signals with magnetoencephalography (MEG) targeting information to obtain targeted specific real-time signals of the activity of emotion-related core brain regions (such as the amygdala).

[0014] The visual feedback module controls the switching of the neural feedback paradigm and displays the guidance and feedback information. It includes a display unit and a paradigm control unit. The specific signals generated by the real-time processing module are encapsulated by the paradigm control unit, and finally the display unit displays the guidance and the encapsulated specific signal feedback information.

[0015] Furthermore, the specific details of the target-specific signal construction method are as follows:

[0016] The system's real-time processing module decodes and analyzes EEG signals and extracts features. Using a multimodal fusion method, it constructs targeted, specific signals for deep brain regions through different decompositions of EEG signals, characterizing the activation status of core areas of the human brain closely related to emotional function.

[0017] Furthermore, the decoding analysis and feature extraction of the electroencephalogram (EEG) signals include the following steps:

[0018] Step (1) EEG signal preprocessing: EEG signals are obtained through signal filtering and channel screening, noise removal and discarding, removal of electrooculography and electromyography, rereference, and independent component analysis;

[0019] Step (2), EEG signal feature extraction: For the EEG signal in step (1), multi-channel and multi-band feature information of the EEG signal is extracted by methods such as information entropy consistency constraint and EEG time-frequency analysis.

[0020] Furthermore, the construction of the specific targeting signal based on EEG-MEG multimodal fusion includes the following steps:

[0021] (1) Synchronous acquisition experiment of EEG / MEG: By conducting a synchronous acquisition experiment of EEG and MEG in a high electromagnetic shielding MEG room, the signal acquisition of physiological paradigms related to emotional faces was carried out to obtain synchronous EEG and MEG data.

[0022] (2) EEG signal feature extraction: The process is the same as the EEG signal decoding process in the above system;

[0023] (3) Preprocessing of magnetoencephalogram (MEG) signals: For the MEG signals in step (1), task-state MEG data are obtained by signal filtering and channel screening, noise removal and discarding, independent component analysis, time-frequency analysis and other methods.

[0024] (4) Extraction of magnetoencephalogram (MEG) signal features: For the MEG signals in step (3), the source signals of whole brain neural activity are estimated by time-frequency analysis and linear constraint minimum variance source reconstruction method.

[0025] (5) Targeted region source signal extraction: For the source signal estimation in step (4), specific neural activity source signals of the emotion-targeted brain region are obtained through analysis methods such as multivariate symmetric orthogonalization and AAL template coordinate ball extraction.

[0026] (6) Construction of specific mapping model: For the EEG multi-band features and the specific neural activity source signals of the emotion-targeting brain regions in steps (2) and (5), a specific mapping model of EEG and EEG is constructed by using convolutional neural networks and other methods, so as to characterize the characteristic activity signals of emotion-related brain regions by cortical EEG signals.

[0027] Furthermore, in step (4), the linear constraint minimum variance source reconstruction method is as follows: Under the condition of satisfying the linear constraint, find W(q0) that minimizes the output variance of the filter, which is the so-called linear constraint minimum variance. Its linear constraint can be expressed as:

[0028] W T (q0)L(q0)=I

[0029] Where q0 represents any position in the source space, W(q0) represents the spatial filter at that position, L(q0) represents the lead matrix at that position in the forward problem, which is related to the neural activity source information and the position of the magnetoencephalogram sensor, and I is the identity matrix;

[0030] In the linearly constrained minimum variance source reconstruction method, the linear constraint ensures that the signal of interest can pass through the filter, while minimizing the variance allows for the optimal allocation of the filter's stopband response, thus minimizing the variance of the output signal. Mathematically, this can be expressed as:

[0031] And W T (q0)L(q0)=I

[0032] tr{} represents the trace, y represents the signal output by the filter, and C(y) represents the variance of y; the above equation can be solved using the Lagrange operator as follows:

[0033] W(q0)=[LT (q0)C -1 L(q0)] -1 L T (q0)C -1 (x)

[0034] Therefore, we can measure the signal x using magnetoencephalography (MEG) cortical electrodes, and estimate the source signal y at any location in the source space using the following formula:

[0035] y = W T (q0)x

[0036] The specific mapping model construction method in step (6) is as follows: using the specific neural activity source signal y of the emotion-targeting brain region obtained in step (5) as the gold standard, and using EEG multi-band multi-channel features as input, regression analysis is performed. The formula is illustrated below:

[0037] y = W * E

[0038] In the formula, E is a 64*8 EEG channel and frequency band feature matrix, y is a specific neural activity source signal, and W represents the regression analysis method. In this invention, a convolutional neural network is used for regression analysis.

[0039] Furthermore, the specific mapping model uses multi-channel, multi-band EEG signals as input and magnetoencephalography (MEG) source-level signals as target signals for regression analysis. A two-dimensional convolutional layer, batch normalization layer, and non-linear activation layer are used as individual convolutional blocks. The neural network consists of multiple convolutional blocks (convolutional layer + normalization layer + non-linear activation layer) and fully connected layers, and a dropout strategy is incorporated to prevent overfitting. The model obtains a specific mapping model of source signals from EEG cortical signals that characterize MEG emotion-targeting brain regions.

[0040] Furthermore, specific neurofeedback methods based on multimodal fusion are applied to guide and optimize anxiety-related neurological symptoms;

[0041] Furthermore, the aforementioned neurofeedback modulation training method requires 4 to 10 consecutive treatment courses; wherein, a single specific neurofeedback training session includes, but is not limited to, the following steps:

[0042] (1) Baseline resting-state EEG: Observe the brain activity of users in the resting state before training, establish the regulatory threshold used in the neurofeedback process, initially set to 90% of the resting-state specific signal energy;

[0043] (2) Practice phase: Help the subjects become familiar with the experimental procedure and assist them in devising 3 to 5 neurofeedback strategies for practice. Select 1 to 2 of them with the best regulatory effect for formal neurofeedback training.

[0044] (3) Feedback training phase: The subjects conducted three sets of formal neurofeedback training using the pre-constructed strategy, and achieved the neurofeedback effect by downregulating the specific signal energy.

[0045] A single training session includes three different scenarios: rest, regulation, and computation.

[0046] In the resting scenario, participants were encouraged to relax as much as possible to allow their neural activity to return to baseline.

[0047] The control scenario provides real-time visual feedback to the subjects, providing them with the intensity of a specific signal. Subjects are required to use feedback strategies to downscale the intensity of this specific signal as much as possible.

[0048] The computational scenario involved participants performing simple mathematical operations to help disrupt and break free from the previously regulated state. Each scenario lasted 40 seconds, and the three scenarios were repeated sequentially a total of four times.

[0049] (4) Transfer training stage: Compared with the feedback training stage, the control scenario no longer provides visual feedback on the specific signal intensity to the subject, prompting the subject to continue subjective control according to the previously constructed strategy, in order to test whether the subject's control ability has been established and can be transferred to ordinary life scenarios.

[0050] The remaining procedures in this phase are the same as those in the feedback training phase.

[0051] (5) Resting-state EEG after training: Tests the changes in brain activity intensity in the subject at rest after a single course of neurofeedback training.

[0052] The beneficial effects of this invention are as follows: This invention proposes a specific signal index combining multimodal fusion for improving a neurofeedback system for anxiety. By synchronously acquiring EEG / MEG data of the human body under emotional facial stimulation, performing EEG / MEG data analysis and decoding, a specific neurofeedback signal guided by MEG is constructed, which has stronger spatial accuracy and symptom specificity. Applying specific signals for neurofeedback therapy allows for the development of multi-course individualized neuromodulation training to improve the user's anxiety. Due to the portability and non-invasiveness of EEG, this invention can perform neuromodulation non-invasively and portablely, and can be easily expanded to community, home and other application scenarios. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the system structure of the specific neurofeedback system for improving anxiety based on multimodal fusion according to the present invention;

[0054] Figure 2 This is a schematic diagram of the technical route for processing electroencephalogram (EEG) data in this invention;

[0055] Figure 3This is a schematic diagram of the technical route for processing magnetoencephalogram (MEG) data in this invention;

[0056] Figure 4 This is a schematic diagram of the technical route for constructing a specific mapping model for MEG-guided EEG in this invention.

[0057] Figure 5 This is a schematic diagram of the convolutional neural network specific mapping model used in a specific embodiment of the present invention;

[0058] Figure 6 This is a schematic diagram of the neural feedback regulation paradigm, method, and steps of the present invention;

[0059] Figure 7 This is a schematic diagram of the training results of a specific embodiment of the specific mapping model of the present invention;

[0060] Figure 8 This is a schematic diagram of the training results of a specific embodiment of the application of the specific mapping model of the present invention to neural feedback training.

[0061] In the diagram, 1 is the EEG acquisition module, 11 is the acquisition unit, and 12 is the communication unit one;

[0062] 2 is the real-time processing module, 21 is the second communication unit, 22 is the data preprocessing module, 23 is the feature extraction unit, and 24 is the multimodal specific signal construction unit.

[0063] 3 is the visual feedback module, 31 is the paradigm control unit, and 32 is the display unit. Detailed Implementation

[0064] To more clearly illustrate the technical solution of the present invention, the present invention will be further described below;

[0065] like Figure 1 As shown, the present invention provides a specific neurofeedback system for improving anxiety based on multimodal fusion, comprising an electroencephalogram (EEG) acquisition module 1, a real-time processing module 2, and a visual feedback module 3.

[0066] The EEG acquisition module 1 is connected to the real-time processing module 2, the real-time processing module 2 is connected to the visual feedback module 3, and the visual feedback module 3 is connected to the EEG acquisition module 1, thus realizing a complete closed-loop treatment system from signal acquisition to signal processing to signal feedback and signal acquisition, helping users establish a platform for autonomous regulation of neural activity.

[0067] The EEG acquisition module 1 includes an acquisition unit 11 and a communication unit 12. The acquisition unit 11 records 64 leads of EEG using an electrode cap. The electrodes also include horizontal electrooculography (EOG), vertical electrooculography (EOG), electromyography (EMG), and electrocardiography (ECG). The electrode layout conforms to the calibrated 10-20 electrode lead positioning standard. During the collection process, the Ref electrode, which is close to the Cz electrode, is used as a reference to keep the electrode impedance ≤10kΩ and the sampling frequency is set to 1000Hz. The communication unit 12 includes software communication processing and hardware communication equipment. It encapsulates and encodes the acquired EEG data and sends it to the communication unit 21 at a rate of 1 frame per second.

[0068] The real-time processing module 2 includes a second communication unit 21, a data preprocessing unit 22, a feature extraction unit 23, and a specific signal construction unit 24. The second communication unit 21 acquires signal frames from the first communication unit 12, decodes and verifies them to obtain the raw EEG signal, and acquires parameters such as the sampling rate, number of channels, and channel position of the current EEG acquisition. The data preprocessing unit 22 preprocesses the acquired EEG data, and the preprocessing process is as follows: Figure 2 As shown in the first part, EEG signal preprocessing is performed through methods including but not limited to signal filtering and channel selection, noise reduction and discarding, removal of electrooculography and electromyography, rereference, and independent component analysis.

[0069] The feature extraction unit 23 further decodes and analyzes the preprocessed EEG data. The feature extraction process is as follows: Figure 2 As shown in the latter part, multi-channel and multi-band feature information of EEG signals is extracted through methods including but not limited to information entropy consistency constraints and EEG time-frequency analysis.

[0070] The multimodal specific signal construction unit 24 adopts, for example, Figure 4 The trained specific mapping model is used to construct specific signals.

[0071] The visual feedback module 3 includes a paradigm control unit 31 and a display unit 32. After acquiring specific signals, the paradigm control unit 31 encapsulates specific signals and guidance messages according to the current neurofeedback training stage, and displays and switches them through the display unit 32. The display unit 32 includes buttons for selecting communication ports, inputting user names and other information, and displays specific signals and guidance messages to help users perform subjective neurofeedback regulation.

[0072] Example 1

[0073] EEG / MEG signals of emotional facial patterns were collected simultaneously from all participants. A specific mapping model targeting deep brain regions (amygdala) using EEG-MEG fusion was constructed, such as... Figure 4 As shown, it includes the following steps:

[0074] Step (1) EEG / MEG synchronous acquisition experiment: EEG and MEG synchronous acquisition experiment was conducted in a highly electromagnetically shielded MEG room to acquire signals of physiological paradigms related to emotions and faces. The CTF275 fully conductive MEG system was used for acquisition, with a sampling frequency of 1200Hz. 64-channel EEG was recorded using an EEG electrode cap, which was fixed with an elastic mesh cap. Ag / AgcL sintered electrode wires were used for the EEG to reduce metallic electromagnetic interference to the MEG room acquisition environment. Head movement interference and electrooculography interference were eliminated before and after acquisition. After the synchronous signal recording was completed, all subjects underwent magnetic resonance imaging (MRI) localization. The scanning parameters were as follows: TR = 1900ms, TE = 2.48ms, FA = 9°, slice number = 176, slice thickness = 1mm, voxel size = 1×1×1mm3, FOV = 250×250mm2. Head movement interference and electrooculography interference were eliminated before and after acquisition. Synchronous EEG and MEG data were obtained.

[0075] Step (2), EEG signal decoding analysis and feature extraction: The process is the same as the EEG signal decoding process in the system described above;

[0076] Step (3), preprocessing of magnetoencephalogram (MEG) signals: such as Figure 3 As shown in the upper part, for the EEG signal in step (1), task-state EEG data are obtained by signal filtering and channel screening, noise removal and discarding, independent component analysis, time-frequency analysis and other methods.

[0077] Step (4), MEG signal feature extraction: such as Figure 3 As shown in the lower part, for the brain magnetic signals in step (3), the source signals of whole brain neural activity are estimated by time-frequency analysis and linear constraint minimum variance source reconstruction method.

[0078] Step (5), Targeted region source signal extraction: such as Figure 3 As shown in the lower part, for the source signal estimation in step (4), specific neural activity source signals of the emotion-targeting brain region (amygdala) are obtained by using multivariate symmetric orthogonalized AAL template coordinate digging and other analysis methods.

[0079] Step (6), Specific mapping model construction: such as Figure 4 As shown, the specific neural activity source signals of the EEG multi-band features and the EEG emotion-targeting brain regions in steps (2) and (5) are used to construct a specific mapping model of EEG and EEG using the convolutional neural network method, so as to characterize the activity signals of the deep emotional brain region (amygdala) with cortical EEG signals.

[0080] In this embodiment, the specific mapping model is as follows: Figure 5As shown, multi-channel, multi-band EEG signals were used as input, and the target signal was the source-level signal from magnetoencephalography (MEG) for regression analysis. A two-dimensional convolutional layer, a batch normalization layer, and a non-linear activation layer were used as individual convolutional blocks. The neural network consisted of four convolutional blocks (convolutional layer + normalization layer + non-linear activation layer) and two fully connected layers, with a dropout strategy to prevent overfitting. Input data was augmented by random sampling and averaging. The model accuracy was evaluated using the R-squared index to assess the similarity between the regression signal and the source signal.

[0081] R-squared calculation formula:

[0082]

[0083] After training with 360,000 samples and testing on a 90,000-sample test set, the results are as follows: Figure 7 As shown, the model fit stabilized at approximately 0.78, and the regression signal could reconstruct the specific neural activity source signals of the emotion-targeting brain region relatively well.

[0084] Example 2

[0085] Furthermore, the aforementioned specific mapping model targeting deep brain regions was applied to a neurofeedback training scenario, selecting healthy subjects with no history of mental illness to undergo neurofeedback training to improve anxiety; for example... Figure 6 As shown, the neurofeedback modulation paradigm described in this embodiment includes the following steps:

[0086] Step (1) Baseline resting-state EEG: Observe the brain activity of the user in the resting state before training, initially set to 90% of the resting-state specific signal energy;

[0087] Step (2), Practice phase: Help the subjects become familiar with the experimental procedure and assist them in devising 3-5 neurofeedback strategies for practice. Select the 1-2 strategies with the best results for formal neurofeedback training.

[0088] Step (3), Feedback Training Phase: Subjects undergo three sets of formal neurofeedback training using pre-constructed strategies, achieving the neurofeedback effect by downregulating specific signal energy; each training session includes three different scenarios: rest, regulation, and computation; in the rest scenario, subjects relax as much as possible to restore neural activity to baseline; in the regulation scenario, specific signal intensity is fed back to subjects in real time via visual feedback, requiring subjects to use feedback strategies to downregulate the specific signal intensity as much as possible; in the computation scenario, subjects perform simple mathematical operations to help disrupt and escape the aforementioned regulation state; each scenario lasts for 40 seconds, and the three scenarios are repeated sequentially four times.

[0089] Step (4), Transfer Training Stage: Compared with the feedback training stage, the control scenario no longer provides visual feedback on the specific signal intensity to the subject, prompting the subject to continue subjective control according to the previously constructed strategy, in order to test whether the subject's control ability has been established and can be transferred to ordinary life scenarios; the rest of the process in this stage is the same as the feedback training stage.

[0090] Step (5), Post-training resting-state EEG: Test the changes in brain activity intensity in the subject at rest after a single course of neurofeedback training.

[0091] In this specific implementation case, through a single treatment course, the difference in specific signal energy of the subjects is as follows: Figure 8 As shown in the figure, after one course of neurofeedback training, the energy of the specific signals related to anxiety decreased, and the subjects maintained their corresponding emotional regulation ability and kept the energy of the relevant signals low even during the transfer training phase. The results indicate that the specific neurofeedback signals we designed can effectively guide individuals in emotional regulation and have targeted training value.

[0092] In summary, this invention presents a specific neurofeedback system for improving anxiety based on multimodal fusion. It simultaneously acquires EEG / MEG data from the human body in response to emotional facial stimuli, analyzes and decodes this data, and constructs a specific neurofeedback signal guided by MEG for EEG, exhibiting stronger spatial accuracy and symptom specificity. Applying this specific signal for neurofeedback therapy allows for the development of multi-course adaptive neuromodulation training to improve the user's anxiety, including but not limited to improvements in anxiety-related scale scores, physiological anxiety indicators, and brain imaging. Due to the portability and non-invasiveness of EEG, this invention significantly reduces patient discomfort, enabling convenient and painless neuromodulation, and can be further developed for home and community settings.

[0093] Finally, it should be understood that the embodiments described in this invention are only used to illustrate the principles of the embodiments of this invention; other variations may also fall within the scope of this invention; therefore, as examples rather than limitations, alternative configurations of the embodiments of this invention can be regarded as consistent with the teachings of this invention; correspondingly, the embodiments of this invention are not limited to the embodiments explicitly introduced and described in this invention.

Claims

1. A specific neurofeedback system for improving anxiety based on multimodal fusion, characterized in that, It includes an interconnected EEG acquisition module (1), a real-time processing module (2), and a visual feedback module (3). The EEG acquisition module (1) is connected to the real-time processing module (2), the real-time processing module (2) is connected to the visual feedback module (3), and the visual feedback module (3) is connected to the EEG acquisition module (1), thus forming a closed-loop system; The electroencephalogram (EEG) acquisition module (1) includes an acquisition unit (11) and a communication unit (12) connected to each other. The communication unit (12) is connected to the communication unit (21) in the real-time processing module (2). The real-time processing module (2) includes a second communication unit (21), a data preprocessing module (22), a feature extraction unit (23), and a multimodal specific signal construction unit (24) that are interconnected. The multimodal specific signal construction unit (24) is connected to the paradigm control unit (31) in the visual feedback module (3). The visual feedback module (3) includes a paradigm control unit (31) and a display unit (32); In the electroencephalogram (EEG) acquisition module (1), the acquisition unit (11) acquires and encapsulates the EEG signals and sends them to the real-time processing module (2) through the communication unit (12); After the communication unit 2 (21) in the real-time processing module (2) acquires EEG data from the EEG acquisition module (1), it first preprocesses the EEG signal through the data preprocessing module (22), then decodes it through the feature extraction unit (23), and through the multimodal specific signal construction unit (24), the decoded signal is passed through the multimodal specific mapping model to obtain the real-time specific mapping signal of the core brain region activity related to emotion. The specific mapping signal generated by the real-time processing module (2) is encapsulated by the paradigm control unit (31) in the visual feedback module (3), and finally the display unit (32) displays the guidance and the encapsulated specific signal feedback information. The specific steps for constructing the specific mapping model are as follows: (1) Synchronous acquisition experiment of EEG / MEG: By conducting a synchronous acquisition experiment of EEG and MEG in a high electromagnetic shielding MEG room, the signal acquisition of physiological paradigms related to emotional faces was carried out to obtain synchronous EEG and MEG data. (2) EEG signal feature extraction: Multi-channel and multi-band feature information of EEG signals are extracted by means of information entropy consistency constraint and EEG time-frequency analysis. (3) Preprocessing of magnetoencephalogram (MEG) signals: For the MEG signals in step (1), task-state MEG data are obtained by signal filtering and channel screening, noise removal and discarding, independent component analysis and time-frequency analysis. (4) Extraction of magnetoencephalogram (MEG) signal features: For the MEG signals in step (3), the source signals of whole brain neural activity are estimated by time-frequency analysis and linear constraint minimum variance source reconstruction method. (5) Targeted region source signal extraction: For the source signal estimation in step (4), the specific neural activity source signal of the emotion-targeting brain region is obtained by multivariate symmetric orthogonalization and AAL template coordinate ball-digging analysis method. (6) Construction of specific mapping model: For the EEG multi-band features and the specific neural activity source signals of the emotion-targeting brain regions in steps (2) and (5), a specific mapping model of EEG and EEG is constructed by convolutional neural network, so as to represent the characteristic activity signals of emotion-related brain regions by cortical EEG signals. Specifically, the linearly constrained minimum variance source reconstruction method is as follows: Under the condition of satisfying the linear constraint, find W(q0) that minimizes the output variance of the filter, which is the so-called linearly constrained minimum variance. Its linear constraint can be expressed as: W T (q0)L(q0)=I In the formula, q0 represents any position in the source space, W(q0) represents the spatial filter at that position, L(q0) represents the lead matrix at that position in the forward problem, which is related to the neural activity source information and the position of the magnetoencephalogram sensor, and I is the identity matrix; In the linearly constrained minimum variance source reconstruction method, the linear constraint ensures that the signal of interest can pass through the filter while minimizing the variance. The optimal allocation of the filter's stopband response minimizes the variance of the output signal, which can be mathematically expressed as: And W T (q0)L(q0)=I In the formula, tr{} represents the trace, y represents the signal output by the filter, and C(y) represents the variance of y; The above equation is solved using the Lagrange operator as follows: W(q0)=[L T (q0)C -1 L(q0)] -1 L T (q0)C -1 (x) Therefore, the source-estimated signal y at any location in the source space, obtained by measuring signal x using magnetoencephalography (MEG) cortical electrodes, is calculated using the following formula: y=W T (q0)x The specific mapping model construction method in step (6) is as follows: using the specific neural activity source signal y of the emotion-targeting brain region obtained in step (5) as the gold standard, and using EEG multi-band multi-channel features as input, regression analysis is performed. The formula is illustrated below: y = W * E In the formula, E is a 64*8 EEG channel and frequency band feature matrix, y is the specific neural activity source signal, and W represents the regression analysis method.

2. The specific neurofeedback system for improving anxiety based on multimodal fusion according to claim 1, characterized in that, The closed-loop system described herein is used to assist in developing individualized neurofeedback modulation training. The neurofeedback mentioned above is a neuromodulation method that requires 4 to 10 consecutive courses of feedback training. The single-session specific neurofeedback training is controlled by the paradigm control unit (31) within the visual feedback module (3), and its specific operation steps are as follows: (1) Baseline resting-state EEG: Observe the brain activity of users in the resting state before training and establish the regulatory threshold used in the neurofeedback process; (2) Practice phase: Help the subjects become familiar with the experimental procedure and assist them in devising 3 to 5 neurofeedback strategies for practice. Based on the regulatory effect, select 1 to 2 of them for formal neurofeedback training. (3) Feedback training phase: The subjects conducted three sets of formal neurofeedback training using the pre-constructed strategy, and achieved the neurofeedback effect by downregulating the specific signal energy. The single training session includes three different scenarios: rest, regulation, and computation. In the resting setting, the subjects relaxed, allowing their neural activity to return to baseline. The control scenario provides real-time visual feedback to the participants, who are then required to use a feedback strategy to downscale the specific signal. In a computational scenario, participants performed simple mathematical operations to help disrupt and break free from the controlled state. Each scene lasts 40 seconds, and the three scenes are repeated in sequence a total of 4 times. (4) Transfer training stage: Compared with the feedback training stage, the control scenario no longer provides visual feedback on the specific signal intensity to the subject, prompting the subject to continue subjective control according to the previously constructed strategy, in order to test whether the subject's control ability has been established and can be transferred to ordinary life scenarios; the rest of the process in this stage is the same as the feedback training stage. (5) Resting-state EEG after training: Tests the changes in brain activity intensity in the subject at rest after a single course of neurofeedback training.

3. The specific neurofeedback system for improving anxiety based on multimodal fusion according to claim 1, characterized in that: The specific features of the convolutional neural network method are as follows: it uses multi-channel, multi-band EEG signals as input and magnetoencephalography (MEG) source-level signals as target signals for regression analysis. The neural network consists of multiple convolutional blocks and fully connected layers, with each two-dimensional convolutional layer, batch normalization layer, and nonlinear activation layer forming a single convolutional block. A random dropout strategy is added to prevent the model from overfitting. The model obtains specific mapping signals that represent specific neural activity source signals of magnetoencephalograms from EEG cortical signals.

Images