A consciousness disorder stimulation regulation system and method fusing electroencephalogram connection recognition

Abstract

The application discloses a kind of consciousness disorder stimulation regulation systems and methods of fusion electroencephalogram connection identification, including simulation electroencephalogram signal data acquisition stage, connection identification analysis stage, stimulation parameter optimization stage and executable stimulation instruction conversion stage;The application has the following advantages and effects: generate whole brain electrical activity distribution map and identify key connection area by simulation electroencephalogram signal data acquisition stage, generate brain function connection atlas in connection identification analysis stage using functional connection analysis network and phase synchronization algorithm, establish the space-time association of whole brain electrical activity distribution map and brain function connection atlas in stimulation parameter optimization stage and generate optimized stimulation parameter set by fusion network using multi-objective optimization algorithm, finally executable stimulation instruction is converted based on adaptive control model in executable stimulation instruction conversion stage, so as to significantly improve the precision and adaptive ability of electroencephalogram signal stimulation regulation.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal processing and neuromodulation technology, and in particular to a system and method for regulating consciousness disorders by integrating EEG connectivity recognition. Background Technology

[0002] In the field of EEG signal processing and neuromodulation, existing technologies typically rely on separate EEG signal analysis and stimulation parameter optimization modules. EEG signal analysis often focuses on extracting local activity features or employing simple functional connectivity calculation methods, while stimulation parameter optimization is based on fixed rules or empirical adjustments, lacking a comprehensive consideration of the spatiotemporal dynamics of EEG signals and the overall characteristics of the functional connectivity network. This separate processing approach makes it difficult to effectively coordinate EEG activity distribution and functional connectivity information, resulting in insufficient precision and adaptability of stimulation modulation. Especially in scenarios requiring high-precision real-time modulation, existing technologies cannot fully utilize the correlation between the multi-scale features of EEG signals (such as micro-transient features, meso-modal features, and macro-trend features) and functional connectivity strength, leading to a mismatch between stimulation parameter settings and the actual EEG state, thus limiting the maximization of modulation effects. Therefore, designing a method that can effectively integrate EEG connectivity recognition and stimulation parameter optimization to achieve collaborative analysis of EEG activity distribution maps and functional connectivity maps has become a core technical problem urgently needing to be solved in this field. Summary of the Invention

[0003] The purpose of this invention is to provide a system and method for regulating consciousness disorders by integrating brainwave connectivity recognition, in order to solve the problems mentioned in the background art.

[0004] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0005] This invention provides a method for modulating consciousness disorders by integrating brainwave connectivity recognition, comprising the following steps:

[0006] Simulated EEG signal data acquisition stage: Simulated EEG signal data is generated based on the EEG signal generation model. The EEG signal generation model is a computational model that integrates the functional region structure defined by the standard brain functional atlas with the neuronal population dynamics described by the neuronal mass differential equation. Spatiotemporal filtering is applied to the simulated EEG signal data to generate a whole brain electrical activity distribution map, and key connectivity regions are identified in the whole brain electrical activity distribution map.

[0007] Connectivity identification and analysis phase: Based on the identified key connectivity regions, a functional connectivity analysis network is deployed, and the connectivity strength between simulated EEG signal data is analyzed using a phase synchronization algorithm to generate a brain functional connectivity map;

[0008] Stimulation parameter optimization stage: Establish the spatiotemporal correlation between the whole brain electrical activity distribution map and the brain functional connectivity map, align the stimulation parameters using a multi-objective optimization algorithm, and integrate the whole brain electrical activity distribution map and the brain functional connectivity map through a fusion network to generate an optimized stimulation parameter set;

[0009] Executable stimulus instruction conversion stage: Based on the adaptive control model, the optimized stimulus parameter set is converted into executable stimulus instructions.

[0010] By employing the above technical solutions, in the simulated EEG signal data acquisition stage, simulated EEG signal data is generated based on the EEG signal generation model, and spatiotemporal filtering is applied to generate a whole-brain EEG activity distribution map, thereby identifying key connectivity regions and providing basic EEG activity data for subsequent analysis. In the connectivity identification and analysis stage, a brain functional connectivity atlas is generated by deploying a functional connectivity analysis network and a phase synchronization algorithm, which can accurately analyze the functional connectivity strength between brain regions and reveal network abnormalities under consciousness impairment. In the stimulation parameter optimization stage, a spatiotemporal correlation is established between the whole-brain EEG activity distribution map and the brain functional connectivity atlas, and a multi-objective optimization algorithm is used to align the stimulation parameters. By integrating multi-source information through a fusion network to generate an optimized set of stimulation parameters, the matching between stimulation parameters and EEG dynamic states is ensured, improving the targeting and efficiency of regulation. Finally, in the executable stimulus instruction conversion stage, the optimized set of stimulation parameters is transformed into executable stimulus instructions based on an adaptive control model, achieving seamless conversion from parameters to instructions and enhancing the practicality and automation of the method. Overall, this invention effectively integrates EEG connectivity recognition and stimulation parameter optimization through multi-stage collaborative processing, overcoming the shortcomings of the separation of EEG signal analysis and stimulation regulation in existing technologies, and significantly improving the accuracy, adaptability, and effectiveness of stimulation regulation for consciousness disorders.

[0011] A further setting is that, during the simulated EEG signal data acquisition phase:

[0012] The standard brain functional atlas predefines key functional network regions, including the default mode network region, the executive control network region, and the salience network region; the neuronal quality differential equation generates EEG waveforms containing different frequency bands by simulating the average membrane potential and synaptic transmission dynamics of a neuronal population.

[0013] By adopting the above technical solution, and by pre-defining key functional network regions, including the default mode network region, executive control network region, and salience network region, the EEG signal generation model can accurately simulate the core brain network activities related to consciousness. Simultaneously, by simulating the average membrane potential and synaptic transmission dynamics of neuronal populations through neuronal mass differential equations, EEG waveforms containing different frequency bands are generated. This makes the simulated EEG signal data closer to the physiological characteristics of real EEG, enhancing the data's authenticity and diversity. This provides richer and more accurate input for generating whole-brain activity distribution maps and identifying key connectivity regions, thereby improving the reliability and effectiveness of the entire method in the diagnosis and regulation of consciousness disorders.

[0014] A further provision is that, during the simulated EEG signal data acquisition phase, the step of generating simulated EEG signal data includes:

[0015] Based on the standard brain functional atlas, the default mode network region, executive control network region, and salience network region are initialized as high-sensitivity focusing regions, and the remaining brain regions are initialized as low-sensitivity background regions; and a sensitivity function is constructed to divide the EEG monitoring space into high-sensitivity focusing regions and low-sensitivity background regions;

[0016] The simulated EEG signal data is analyzed in real time through a preset anomaly response network. The input to the anomaly response network is the simulated EEG signal data, and the output is transient feature activity areas. The focusing area is dynamically updated based on these transient feature activity areas. If a new transient feature activity area is detected, it is upgraded to a high-sensitivity focusing area. If no transient feature activity is detected in an existing high-sensitivity focusing area, it is downgraded to a low-sensitivity background area. The sampling frequency of the high-sensitivity focusing area is higher than that of the low-sensitivity background area. The simulated EEG signal data is obtained by dynamically adjusting the virtual electrode operating parameters through configuration commands.

[0017] By adopting the above technical solution, the default mode network region, executive control network region, and salience network region are initialized as high-sensitivity focusing areas based on standard brain functional atlases, while the remaining brain regions are initialized as low-sensitivity background areas. A sensitivity function is constructed to divide the EEG monitoring space, achieving priority attention and efficient monitoring of key functional network regions. The simulated EEG signal data is analyzed in real time through a preset abnormal response network, outputting transient feature active areas. The focusing area range is dynamically updated based on the transient feature active areas, which can adaptively adjust the monitoring focus to ensure timely capture of dynamic changes in EEG activity. By setting the sampling frequency of the high-sensitivity focusing area to be higher than that of the low-sensitivity background area, and dynamically adjusting the working parameters of the virtual electrodes through configuration commands, the allocation of computing resources is optimized while ensuring data quality, improving the efficiency and accuracy of simulated EEG signal data acquisition. This provides more accurate and timely data support for the subsequent generation of whole EEG activity distribution maps and identification of key connectivity regions.

[0018] A further setting includes, during the simulated EEG signal data acquisition phase, the step of applying spatiotemporal filtering to the simulated EEG signal data to generate a whole-brain activity distribution map, comprising:

[0019] Bandpass filtering and normalization are performed on the simulated EEG signal data, and the normalized simulated EEG signal data is integrated into a multidimensional tensor according to the virtual electrode channel dimension;

[0020] A spatiotemporal coding network architecture is constructed, which includes a forward feature extraction path, a reverse feature reconstruction path, and a cross-scale feature fusion path.

[0021] The integrated multidimensional tensor is input into the spatiotemporal coding network architecture; in the forward feature extraction path, a deep convolutional network is used to compress and abstract the features of the multidimensional tensor to generate a multi-scale feature mapping set, which includes micro-transient features, meso-mode features and macro-trend features.

[0022] In the cross-scale feature fusion path, the micro-transient features and meso-mode features obtained in the forward feature extraction path are respectively passed to the corresponding resolution levels in the reverse feature reconstruction path.

[0023] In the reverse feature reconstruction path, the macro trend features are upsampled to the meso scale through bilinear interpolation and fused with the meso mode features from the cross-scale feature fusion path; then the fused meso features are restored to the micro scale through interpolation and fused with the micro transient features from the cross-scale feature fusion path, finally outputting a high-resolution feature map that combines detail and wholeness.

[0024] The high-resolution feature maps output by the reverse feature reconstruction path are superimposed and reconstructed to generate the whole brain electrical activity distribution map.

[0025] By employing the aforementioned technical solutions, bandpass filtering and normalization are performed on simulated EEG signal data, and the normalized data is integrated into a multidimensional tensor according to the virtual electrode channel dimension, effectively removing noise and standardizing the data format. A spatiotemporal coding network architecture is constructed, including a forward feature extraction path, a reverse feature reconstruction path, and a cross-scale feature fusion path, achieving comprehensive extraction and fusion of multi-scale features of EEG signals. In the forward feature extraction path, a deep convolutional network is used for feature compression and abstraction, generating a multi-scale feature map set including micro-transient features, meso-mode features, and macro-trend features, capable of capturing complete information from local transient activity to global trends. In the cross-scale feature fusion path, micro-transient features and meso-mode features are passed to the reverse feature reconstruction path, ensuring the preservation of detailed features. In the reverse feature reconstruction path, bilinear interpolation upsampling and feature fusion are used to finally output a high-resolution feature map that combines detail and overall representation, which is then overlaid and reconstructed to generate a full EEG activity distribution map, thus providing a more accurate and comprehensive spatial distribution representation of EEG activity and providing high-quality input for key connectivity region identification and subsequent analysis.

[0026] A further provision includes, during the simulated EEG signal data acquisition phase, a step of identifying key connectivity regions in the whole-brain electrical activity distribution map, comprising:

[0027] Based on the whole brain activity distribution map, the connection strength score of each virtual electrode channel is calculated. The connection strength score is obtained by extracting the intensity of the micro transient features, the stability of the meso mode features and the consistency of the macro trend features fused at each spatial location point in the whole brain activity distribution map, and performing nonlinear weighted aggregation.

[0028] A primary threshold and a secondary threshold for connection strength scoring are preset. The connection strength score is compared with both the primary and secondary thresholds. If the connection strength score is lower than the primary threshold, the corresponding channel is marked as a low connection domain. If the connection strength score is higher than the primary threshold but lower than the secondary threshold, the corresponding channel is marked as a medium connection domain. If the connection strength score is higher than the secondary threshold, the corresponding channel is marked as a high connection domain.

[0029] Different color codes were used to identify key connectivity regions in the whole brain activity distribution map. Different colors corresponded to different connectivity strength levels. Low connectivity regions were defined as weak connectivity regions, medium connectivity regions as medium connectivity regions, and high connectivity regions as strong connectivity regions.

[0030] By employing the aforementioned technical solution, the connectivity strength score of each virtual electrode channel is calculated based on the whole-brain electroencephalogram (EEG) activity distribution map. Nonlinear weighted aggregation of the intensity of micro-transient features, the stability of meso-mode features, and the consistency of macro-trend features is then performed. This comprehensive assessment of the connectivity importance of each channel avoids bias caused by a single feature. By presetting primary and advanced thresholds for connectivity strength scores and labeling the scores as low, medium, or high connectivity regions, quantitative classification of connectivity regions is achieved. Different color coding is used to identify key connectivity regions, making the connectivity strength levels visual and intuitively displaying the distribution of strong and weak brain functional connectivity. This facilitates the implementation of differentiated strategies for different connectivity regions in subsequent analysis, improving the accuracy and operability of connectivity identification analysis. It also provides clear target areas for the deployment of functional connectivity analysis networks and the generation of brain functional connectivity maps.

[0031] A further setting is that the steps in the connection identification and analysis phase include:

[0032] Based on the identified key connectivity regions, a high-density electrode simulation array is deployed at the corresponding location in the virtual brain model;

[0033] Configure the analysis parameters of the functional connectivity analysis network and implement differentiated connectivity analysis strategies for different connectivity regions: configure a first analysis window for high connectivity regions, a second analysis window for medium connectivity regions, and a third analysis window for low connectivity regions; wherein the duration of the first analysis window is shorter than the duration of the second analysis window, and the duration of the second analysis window is shorter than the duration of the third analysis window.

[0034] Through the aforementioned functional connectivity analysis network, a phase synchronization algorithm is applied to perform functional connectivity analysis on the simulated EEG signal data collected by the high-density electrode simulation array. A multi-band coupled phase correction model is introduced, and the synchronization parameters in the phase correction model are continuously optimized through an iterative phase analysis algorithm. The iteration is terminated when the change in the phase consistency index is lower than the preset convergence threshold.

[0035] A brain functional connectivity matrix is ​​constructed based on graph theory network analysis algorithms. Key connectivity communities are calculated using community detection algorithms, and finally, a brain functional connectivity map with connectivity strength labels is output.

[0036] By employing the aforementioned technical solutions, high-density electrode simulation arrays are deployed at corresponding locations in the virtual brain model based on the identified key connectivity regions, ensuring the targeted nature and coverage of data acquisition. By configuring the analysis parameters of the functional connectivity analysis network and implementing differentiated connectivity analysis strategies for different levels of connectivity regions—including a shorter first analysis window for high-connectivity regions, a second analysis window for medium-connectivity regions, and a longer third analysis window for low-connectivity regions—the temporal dynamic characteristics of different connectivity regions are adapted, optimizing analysis efficiency and accuracy. The functional connectivity analysis network utilizes a phase synchronization algorithm for functional connectivity analysis and introduces a multi-band coupled phase correction model. Iterative phase analysis algorithms are used to optimize synchronization parameters until the change in phase consistency index falls below a preset convergence threshold, improving the accuracy and robustness of functional connectivity calculations. A brain functional connectivity matrix is ​​constructed based on graph theory network analysis algorithms, and key connectivity communities are calculated using community detection algorithms. Finally, a brain functional connectivity map with connectivity strength labels is output, revealing the modular structure and key hubs of the brain functional network, providing a reliable connectivity information foundation for stimulus parameter optimization.

[0037] A further setting includes, in the stimulation parameter optimization phase, the step of establishing the spatiotemporal correlation between the whole-brain electrical activity distribution map and the brain functional connectivity map, comprising:

[0038] A global coordinate system is established using the preset reference points of the virtual brain model;

[0039] The electrode coordinates in the whole brain activity distribution map are mapped to the global coordinate system to obtain brain activity distribution data in the global coordinate system.

[0040] Map the coordinates of the connectivity communities in the brain functional connectivity atlas to the global coordinate system to obtain functional connectivity data in the global coordinate system;

[0041] In the global coordinate system, the EEG activity distribution data and the functional connectivity data are spatially registered to establish a spatial association;

[0042] Based on a unified time reference, the timestamps of the EEG activity distribution data and the functional connectivity data are aligned to establish a time correlation.

[0043] By adopting the above technical solutions, a global coordinate system is established using preset reference points of the virtual brain model, providing a unified spatial reference framework for the whole-brain electroencephalogram (EEG) distribution map and the brain functional connectivity map, ensuring the consistency of spatial positions among the data. Electrode coordinates in the EEG distribution map and connectivity community coordinates in the brain functional connectivity map are mapped to the global coordinate system, obtaining EEG distribution data and functional connectivity data in the global coordinate system, achieving spatial standardization of multi-source data. Spatial registration of EEG distribution data and functional connectivity data is performed in the global coordinate system to establish spatial correlation, ensuring the accuracy of stimulus target localization. The timestamps of EEG distribution data and functional connectivity data are aligned based on a unified time reference to establish temporal correlation, ensuring data synchronization in the time dimension, thereby improving the spatiotemporal accuracy and coordination of stimulus modulation.

[0044] A further step is that, in the stimulation parameter optimization stage, a multi-objective optimization algorithm is used to align the stimulation parameters, and the whole-brain electrical activity distribution map and the brain functional connectivity map are integrated through a fusion network to generate an optimized stimulation parameter set, including:

[0045] The fusion network comprises a data input layer, a feature transformation layer, a cross-modal interaction layer, and a confidence output layer. EEG activity distribution data and functional connectivity data in the global coordinate system are input to the data input layer. In the feature transformation layer, the EEG activity distribution data is converted into activity feature vectors, and the functional connectivity data is converted into connectivity feature vectors. Through the cross-modal interaction layer, the activity feature vectors and connectivity feature vectors are cross-fused to generate a fused feature vector. The fused feature vector is input to the confidence output layer to calculate the initial stimulus parameter set.

[0046] Using the initial stimulus parameter set as input, a multi-objective optimization algorithm is invoked, including connection enhancement objective, state stability objective and resource economy objective as optimization objectives, and constraints including physically immutable constraints and policy adjustable constraints are configured.

[0047] Under the constraints, the multi-objective optimization algorithm evaluates multiple randomly generated stimulus strategy parameter schemes in parallel, calculates the achievement score of each stimulus strategy parameter scheme relative to each optimization objective, and aggregates the achievement scores based on preset weights to obtain the corresponding comprehensive evaluation value of the scheme.

[0048] The stimulus strategy parameter scheme with the highest comprehensive evaluation value is selected from all stimulus strategy parameter schemes. The stimulus frequency, stimulus intensity and stimulus location parameters contained therein are combined and encoded, and finally output as the optimized stimulus parameter set.

[0049] By employing the above technical solution, a fusion network comprising a data input layer, a feature transformation layer, a cross-modal interaction layer, and a confidence output layer is used. After inputting EEG activity distribution data and functional connectivity data in the global coordinate system, these are converted into activity feature vectors and connectivity feature vectors, respectively. Then, the cross-modal interaction layer performs feature cross-fusion to generate a fused feature vector. Finally, the confidence output layer calculates the initial stimulus parameter set, achieving deep integration and efficient utilization of multi-source information. Using the initial stimulus parameter set as input, a multi-objective optimization algorithm is invoked, with connectivity enhancement, state stability, and resource economy objectives as optimization goals, and physical immutability is configured. Variable constraints and policy-adjustable constraints are used as constraints to ensure a balance between enhancing connectivity, maintaining state stability, and conserving resources in stimulus parameter schemes. Under these constraints, multiple stimulus strategy parameter schemes are evaluated in parallel. The achievement scores of each optimization objective are calculated and weighted to obtain a comprehensive evaluation value, thereby scientifically selecting the optimal scheme. Finally, the stimulus frequency, stimulus intensity, and stimulus location parameters of the stimulus strategy parameter scheme with the highest comprehensive evaluation value are combined and encoded to output as an optimized stimulus parameter set. This ensures that the stimulus parameters meet the requirements of multi-objective optimization and adapt to actual constraints, significantly improving the overall performance and practicality of stimulus regulation.

[0050] A further provision is that the steps of the executable stimulus instruction switching phase include:

[0051] The optimized stimulus parameter set is input into the adaptive control model. The adaptive control model compiles the stimulus frequency, stimulus intensity, and stimulus position parameters in the optimized stimulus parameter set into executable stimulus instructions according to a preset instruction mapping protocol. Among them, the stimulus frequency parameter is converted into a periodic timing trigger instruction, the stimulus intensity parameter is quantized into an output amplitude instruction of analog voltage or current, and the stimulus position parameter is converted into an address gating instruction of the target virtual electrode through a spatial encoder.

[0052] By adopting the above technical solution, the optimized stimulus parameter set is input into the adaptive control model, and the stimulus frequency, stimulus intensity, and stimulus location parameters are compiled into executable stimulus commands according to a preset instruction mapping protocol. This achieves efficient conversion from abstract parameters to specific commands. Specifically, the stimulus frequency parameter is converted into periodic timing trigger commands, ensuring the timing accuracy of the stimulus; the stimulus intensity parameter is quantified into output amplitude commands of analog voltage or current, ensuring the controllability of the stimulus intensity; and the stimulus location parameter is converted into address gating commands for the target virtual electrode through a spatial encoder, achieving spatial targeting of the stimulus. This instruction conversion mechanism enables the optimized stimulus parameters to directly drive the stimulus device, improving the automation and real-time performance of the entire method and providing reliable technical support for the precise control of consciousness disorders.

[0053] This invention also provides a consciousness disorder stimulation modulation system integrating EEG connectivity recognition, comprising the following modules:

[0054] The simulated EEG signal data acquisition module is configured to execute the steps of the simulated EEG signal data acquisition phase.

[0055] The connection identification and analysis module is configured to perform the steps of the connection identification and analysis phase.

[0056] The stimulus parameter optimization module is configured to execute the steps of the stimulus parameter optimization phase.

[0057] An executable stimulus instruction conversion module is configured to execute the steps of the stimulus regulation simulation phase.

[0058] In summary, the present invention has the following beneficial effects: by simulating the EEG signal data acquisition stage to generate a whole-brain EEG activity distribution map and identify key connectivity regions, in the connectivity identification and analysis stage, a brain functional connectivity map is generated using a functional connectivity analysis network and a phase synchronization algorithm, in the stimulation parameter optimization stage, the spatiotemporal correlation between the whole-brain EEG activity distribution map and the brain functional connectivity map is established and an optimized stimulation parameter set is generated through a fusion network using a multi-objective optimization algorithm, and finally, in the executable stimulation command conversion stage, it is converted into executable stimulation commands based on an adaptive control model, thereby significantly improving the accuracy and adaptability of EEG signal stimulation modulation. Attached Figure Description

[0059] Figure 1 This is the main flowchart of a method for regulating consciousness disorders by integrating brainwave connectivity recognition;

[0060] Figure 2 This is a flowchart illustrating the simulated EEG signal data acquisition stage in a method for regulating consciousness disorders that integrates EEG connectivity recognition.

[0061] Figure 3 This is a flowchart illustrating the connectivity identification and analysis stage in a method for regulating consciousness disorders that integrates EEG connectivity identification.

[0062] Figure 4 This is a flowchart illustrating the stimulation parameter optimization stage in a method for regulating consciousness disorders that integrates EEG connectivity recognition.

[0063] Figure 5 This is a flowchart illustrating the executable stimulus instruction conversion stage in a method for regulating consciousness disorders that integrates brainwave connectivity recognition.

[0064] Figure 6 This is a schematic diagram of a consciousness disorder stimulation regulation system that integrates brainwave connectivity recognition. Detailed Implementation

[0065] The present invention will be further described in detail below with reference to the accompanying drawings.

[0066] As attached Figures 1-6 As shown;

[0067] This embodiment discloses a method for regulating consciousness disorders by integrating brainwave connectivity recognition, comprising the following steps:

[0068] Simulated EEG signal data acquisition stage: Simulated EEG signal data is generated based on the EEG signal generation model. The EEG signal generation model is a computational model that integrates the functional region structure defined by the standard brain functional atlas with the neuronal population dynamics described by the neuronal mass differential equation. Spatiotemporal filtering is applied to the simulated EEG signal data to generate a whole-brain electrical activity distribution map, and key connectivity regions are identified in the whole-brain electrical activity distribution map.

[0069] The specific implementation process is as follows: Standard EEG functional atlas predefines key functional network regions, including the default mode network region, executive control network region, and salience network region. These regions are spatially divided based on internationally recognized brain anatomical atlases, and their functional characteristics are labeled according to functional magnetic resonance imaging (fMRI) or positron emission tomography (PET) data. The default mode network region mainly covers brain regions such as the posterior cingulate cortex and medial prefrontal cortex, responsible for internal conscious activities; the executive control network region includes brain regions such as the dorsolateral prefrontal cortex and anterior cingulate cortex, involved in cognitive control and decision-making; the salience network region includes brain regions such as the anterior insula and anterior cingulate cortex, used to process salient stimuli and attention switching. The coordinates and boundaries of these key functional network regions are encoded as three-dimensional spatial masks, used to initialize high-sensitivity focusing areas and low-sensitivity background areas in the virtual brain model.

[0070] The neuronal mass differential equation (NMEE) generates EEG waveforms containing different frequency bands (such as delta, theta, alpha, beta, and gamma bands) by simulating the average membrane potential and synaptic transmission dynamics of a neuronal population. The NMEE is based on a population neuron model, treating each functional region as a cluster of neurons whose dynamic behavior is described by a set of coupled ordinary differential equations. Specifically, for each functional region, the average membrane potential... and synaptic current The evolution follows the following equation:

[0071] ;

[0072] ;

[0073] in, Indicates the first Each functional area in time The average membrane potential; Indicates the rate of change of membrane potential; The membrane time constant has a value of 10-20 milliseconds; For film capacitance, the value is 1-2 microfarads / square centimeter; For the first The sum of synaptic currents received by each region; Indicates the rate of change of synaptic current; This is the synaptic time constant, with a value ranging from 2 to 10 milliseconds; To the region To the area The connection weights are assigned based on the structural connectivity strength of the standard EEG functional atlas. The interregional conduction delay depends on the Euclidean distance and the white matter conduction velocity; Represents the delayed membrane potential, representing the first... individual brain regions in time The membrane potential; Let be the sigmoid activation function, representing the nonlinear response of synaptic transmission, and its form is: ,in This is the slope parameter (values ​​range from 0.5 to 1.0). The threshold parameter is set to 5-10 mV. Solving the above equations using numerical integration methods (such as the fourth-order Runge-Kutta method) generates multi-channel simulated EEG signal data. The waveforms exhibit oscillation characteristics across different frequency bands: the delta band (1-4 Hz) is achieved by adjusting slow synaptic current parameters; the theta band (4-8 Hz) is generated by hippocampal-cortical circuit dynamics; the alpha band (8-13 Hz) is associated with thalamo-cortical feedback; the beta band (13-30 Hz) originates from local inhibitory interactions; and the gamma band (30-100 Hz) is simulated through rapid excitatory synaptic transmission. The superposition of these frequency band waveforms forms a realistic simulated EEG signal, providing fundamental data input for subsequent spatiotemporal filtering and identification of key connectivity regions.

[0074] The specific implementation process for generating simulated EEG signal data is as follows: First, based on a standard brain functional atlas, the default mode network region, executive control network region, and salience network region are initialized as high-sensitivity focusing areas, while the remaining brain regions are initialized as low-sensitivity background areas. Sensitivity function The construction is achieved through a space weighting function, which takes the following form:

[0075] ;

[0076] in, The spatial coordinates in the virtual brain model are represented by a weight of 1 for the high-sensitivity focusing area and 0.2 for the low-sensitivity background area, thus dividing the EEG monitoring space into differentiated sensitivity regions. Subsequently, a pre-defined anomaly response network analyzes the simulated EEG signal data in real time. The anomaly response network employs a deep convolutional neural network architecture, and its input is multi-channel simulated EEG signal data (dimension...). ,in This represents the number of virtual electrode channels. The time series length is given, and the output is a spatial distribution map of transient feature activity regions. The anomaly response network consists of three core components: a feature extraction layer, an anomaly detection layer, and a spatial mapping layer. The feature extraction layer uses a one-dimensional convolutional kernel (size...). Step 1) Perform sliding window analysis on the input signal to extract time-domain features; the anomaly detection layer captures time dependencies through a gated loop unit module and calculates the anomaly probability at each time point; the spatial mapping layer associates the anomaly probability with the virtual electrode coordinates to generate a binary mask of the transient feature active region (where a value of 1 represents an active region and a value of 0 represents an inactive region). The logic for dynamically updating the focusing area range is as follows: periodically (e.g., every 100 milliseconds) check the status of the transient feature active region. If a new transient feature active region is detected (i.e., the region was not marked as a high-sensitivity focusing region in the previous cycle, but the anomaly probability in the current cycle exceeds the threshold), the layer will detect the anomaly. If the sensitivity is high, then the region is upgraded to a high-sensitivity focusing area, and the sensitivity function is updated. If no transient activity is detected in the original high-sensitivity focusing area (i.e., the probability of anomalies is below the threshold for multiple consecutive cycles); If the area is not specified, it will be downgraded to a low-sensitivity background area. The sampling frequency of the high-sensitivity focusing area is set to... This ensures the capture of high-frequency transient signals, while the sampling frequency in the low-sensitivity background region is [missing information]. To reduce computational load, the virtual electrode operating parameters are dynamically adjusted via configuration commands to obtain simulated EEG signal data. The configuration commands are encapsulated in JSON format and include parameters such as electrode gain, filter cutoff frequency, and sampling mode. For high-sensitivity focusing areas, the electrode gain is set to... The bandpass filter range is For the low-sensitivity background region, the electrode gain is The bandpass filter range is The virtual electrode operating parameters are adjusted in real time via a digital-to-analog converter module to generate multi-channel simulated EEG signal data. The entire process is achieved through closed-loop control, ensuring that the simulated EEG signal data reflects whole-brain activity while prioritizing the capture of dynamic changes in key functional networks.

[0077] The specific implementation process for generating a whole-brain activity distribution map by applying spatiotemporal filtering to simulated EEG signal data is as follows: First, bandpass filtering and normalization are performed on the simulated EEG signal data. A fourth-order Butterworth filter is used for bandpass filtering, with a passband frequency range of 1-100Hz to cover typical EEG signal frequency bands (including delta, theta, alpha, beta, and gamma bands), while suppressing low-frequency drift and high-frequency noise. The filtered signal is then normalized to eliminate inter-channel amplitude differences and baseline fluctuations. The normalized simulated EEG signal data is integrated into a multidimensional tensor according to the virtual electrode channel dimension. The tensor dimension is... (in This represents the total number of virtual electrode channels. (where the time series length is ), and further converted to a time-frequency representation using a short-time Fourier transform, expanding it to... dimensional tensor ( (Indicates the number of frequency points) to capture the frequency domain characteristics of the signal.

[0078] Subsequently, a spatiotemporal coding network architecture was constructed, consisting of a forward feature extraction path, a backward feature reconstruction path, and a cross-scale feature fusion path. The forward feature extraction path employs a deep convolutional network, with the aforementioned multidimensional tensor as its input. The deep convolutional network contains multiple convolutional blocks, each consisting of a convolutional layer, a batch normalization layer, and a ReLU activation function. The kernel size is set to... The step size is 1, and the fill method is "same fill" to maintain spatial resolution. The pooling layer uses max pooling, and the pooling window is [value missing]. The step size is 2, used for progressive downsampling. Through forward propagation, the network generates a multi-scale feature map set: the shallow layer output retains high spatial resolution, capturing micro-transient features (such as high-frequency oscillations and transient events); the middle layer output undergoes two downsampling iterations, capturing meso-level pattern features (such as rhythm synchronization and pattern repetition); the deep layer output undergoes four downsampling iterations, capturing macro-level trend features (such as global activity trends and slow-wave oscillations). The dimension of the micro-transient features is... The dimensions of the meso-level model characteristics are The dimensions of macro trend characteristics are ;in, and For spatial dimensions, , and These represent the number of channels for micro-transient feature mapping, meso-mode feature mapping, and macro-trend feature mapping, respectively. In the cross-scale feature fusion path, the micro-transient features and meso-mode features obtained in the forward feature extraction path are passed to the corresponding resolution levels in the reverse feature reconstruction path via skip connections. Specifically, micro-transient features are directly passed to the input level of the reverse reconstruction path, while meso-mode features are passed to intermediate levels to preserve multi-scale information flow. Cross-scale fusion employs a channel attention mechanism, calculating the weights of each channel through global average pooling and fully connected layers to enhance the contribution of important features. In the reverse feature reconstruction path, the macro-trend features are first upsampled to the meso-scale using bilinear interpolation. Bilinear interpolation is based on a weighted average of the four nearest neighbors, calculated using the following formula: ;in, Indicates position coordinates The upsampled feature values ​​obtained by bilinear interpolation are used to restore the low-resolution feature map to a high resolution. Neighboring points Eigenvalues ​​at; The weights are determined by the distance between the interpolation point and its neighbors. The upsampled macroscopic trend features are fused with the mesoscopic pattern features from the cross-scale feature fusion path, using an element-wise addition method: ;in, This represents the fused mesoscale feature mapping, which combines macroscopic trend features and mesoscale pattern features after upsampling; This indicates the macroscopic trend characteristics after upsampling via bilinear interpolation; This represents the mesoscale mode features derived from the cross-scale feature fusion path. Furthermore, the fused mesoscale features are upsampled again using bilinear interpolation to recover the microscale, and then fused with the microscale transient features from the cross-scale feature fusion path. ;in, This represents the fused high-resolution feature map; This represents the mesoscopic features after upsampling via bilinear interpolation, i.e. The results after upsampling; This represents the microscopic transient features derived from the cross-scale feature fusion path. The final output is a high-resolution feature map with dimensions consistent with the original input spatial resolution (i.e., ...). , (The number of channels for the final high-resolution feature map output), which combines detailed local information and global context.

[0079] Finally, the high-resolution feature maps output by the reverse feature reconstruction path are overlaid and reconstructed to generate a whole-brain electrical activity distribution map. Overlay reconstruction is achieved through... The convolutional layer compresses multi-channel feature maps into single-channel activation maps, which are then normalized to the [0,1] interval using the Sigmoid function to represent the intensity of EEG activity at each spatial location. The whole-brain EEG activity distribution map is presented as a two-dimensional heatmap, where each pixel corresponds to the standardized activity value of the virtual electrode channel, facilitating intuitive assessment of whole-brain dynamics and subsequent identification of key connectivity regions.

[0080] The specific implementation process for identifying key connectivity regions in a whole-brain electroencephalogram (EEG) is as follows: Based on the EEG, a connectivity strength score is calculated for each virtual electrode channel. This score is obtained by extracting the intensity of micro-transient features, the stability of meso-mode features, and the consistency of macro-trend features fused at each spatial point in the EEG, and then performing nonlinear weighted aggregation. Specifically, for each virtual electrode channel, its connectivity strength score... The calculation formula is: ;in, The intensity of the micro-transient feature is represented by the L2 norm of the micro-transient feature vector, i.e. ,in This is the micro-transient feature vector extracted from the whole brain electrical activity distribution map. The stability of the meso-level model characteristics is represented by calculating the reciprocal of the variance of this characteristic within a sliding time window (with a window length of 100 milliseconds). ;in, This represents the feature vector of the mesoscopic model. express The variance is used to quantify the degree of fluctuation of meso-mode features within a sliding time window (e.g., a window length of 100 milliseconds); For a small constant (with a value of...) To prevent division by zero errors, the smaller the variance, the higher the stability. The consistency of macroeconomic trend characteristics is indicated by calculating the Pearson correlation coefficient between this macroeconomic trend characteristic and the global average trend, i.e. ;in, This represents the macroeconomic trend feature vector. This is the global average trend vector; This represents the Pearson correlation coefficient calculation function, used to measure... and The degree of linear correlation between them. The weighting coefficients for micro-level transient characteristics, meso-level pattern characteristics, and macro-level trend characteristics are respectively, satisfying... And dynamically adjust according to the current type of consciousness impairment. For example, during the consciousness recovery period, prioritize increasing the weight of micro-transient features (assuming...). For periods of stable consciousness, the focus is on the consistency of macroeconomic trend characteristics (assuming...). ). This is the Sigmoid function, used to compress the score to the [0,1] interval, enhancing the non-linear discriminative ability of the score.

[0081] Preset connection strength score initial threshold Advanced threshold for connection strength score ;in The value is 0.3. The value is 0.7; the initial threshold for connection strength scoring. Advanced threshold for connection strength score Based on the statistical distribution of connectivity strength in historical EEG data, and using cluster analysis (such as the K-means algorithm) to divide a large number of normal and abnormal EEG samples into connectivity strength scores, the objectivity of the threshold setting is ensured. The connectivity strength score for each virtual electrode channel is then calculated. respectively with and Compare: If If so, then mark the channel as a low-connection domain; if If, then it is marked as a join field; if High connectivity regions are then marked as such. Different color codes are used to identify key connectivity regions in the whole-brain electroencephalogram (EEG) activity distribution map: low connectivity regions are defined as weakly connected regions, using blue coding (RGB values ​​0, 0, 255); medium connectivity regions are defined as moderately connected regions, using yellow coding (RGB values ​​255, 255, 0); and high connectivity regions are defined as strongly connected regions, using red coding (RGB values ​​255, 0, 0). The coded connectivity regions are mapped onto the spatial coordinates of the EEG activity distribution map using color overlay technology, generating a visual connectivity strength heatmap. The color transparency is dynamically adjusted according to the connectivity strength score; the higher the score, the more saturated the color, facilitating intuitive identification of key connectivity regions and providing a spatial localization basis for subsequent functional connectivity analysis.

[0082] Connection identification and analysis phase: Based on the identified key connection regions, a functional connectivity analysis network is deployed, and the connection strength between simulated EEG signal data is analyzed through a phase synchronization algorithm to generate a brain functional connectivity map.

[0083] The specific implementation process is as follows: Based on the identified key connectivity regions, a high-density electrode simulation array is deployed at the corresponding location in the virtual brain model. The high-density electrode simulation array is positioned based on the spatial coordinates of the standard brain functional atlas to ensure that the electrodes cover all key functional network regions, such as the default mode network region, the executive control network region, and the salience network region. The high-density electrode simulation array consists of virtual electrode nodes, each of which is equipped with an independent signal acquisition module, capable of capturing multi-channel time series of simulated EEG signal data in real time. The functional connectivity analysis network (FADA) is configured with different analysis parameters, implementing differentiated connectivity analysis strategies for different connectivity regions: a first analysis window of 100 ms is configured for high connectivity regions to capture rapidly changing transient connectivity patterns; a second analysis window of 500 ms is configured for medium connectivity regions to balance temporal resolution and stability; and a third analysis window of 1000 ms is configured for low connectivity regions to accumulate sufficient data for low-frequency connectivity analysis. The duration of the first analysis window is shorter than that of the second, and the duration of the second is shorter than that of the third. This differentiated design ensures efficient allocation of analysis resources, prioritizing the dynamic changes in high connectivity regions. Using the FADA network, a phase synchronization algorithm is applied to perform functional connectivity analysis on simulated EEG signal data acquired by a high-density electrode array. The phase synchronization algorithm extracts the instantaneous phase of each channel signal based on the Hilbert transform and calculates the phase lock value between channels, using the following formula: ;in, This is the phase-locked value, and its range is: The closer the value is to 1, the more stable the phase relationship between the two channels is during the observation period, and the stronger the synchronization; the closer the value is to 0, the more random the phase relationship is, and the weaker the synchronization. For modulo operation; For time points; The imaginary unit satisfies ; is the base of the natural logarithm; and They represent channels respectively. and In time The instantaneous phase. A multi-band coupled phase correction model is introduced, which integrates the phase synchronization contributions from different frequency bands (such as delta, theta, alpha, beta, and gamma) through weighted integration. Its core equation is: ;in, This is the total phase lock value; Represents a set of frequency bands; For frequency band index variables; The dynamic weights for each frequency band are adaptively adjusted based on the current state of consciousness. Indicates a specific frequency band The phase lock value is calculated above. The synchronization parameters in the phase correction model, including weights, are continuously optimized using an iterative phase analysis algorithm. The iterative process, including a phase deviation compensation term, minimizes the phase residual based on gradient descent. Iteration terminates when the phase consistency index (such as the change in the global phase consistency index) falls below a preset convergence threshold (0.01), ensuring the model converges to a stable state. A brain functional connectivity matrix is ​​constructed based on graph theory network analysis algorithms, with virtual electrodes as nodes. The values ​​are edge weights, forming a weighted undirected graph; key connection communities are calculated using community detection algorithms (such as the Louvain method) to identify functional modular structures. Community partitioning is based on the principle of maximizing modularity. The calculation formula is: ;in, is the normalization factor, where This represents the total number of edges in the network. To connect matrix elements, representing nodes and nodes The actual strength of the connection between them; Represents a node The degree of the node; Represents a node The degree of the node; Represents a node and nodes The expected connection strength between them; Kronek Function, when node and nodes They were assigned to the same community (i.e.) When they belong to different communities, their value is 1; when they belong to different communities (i.e., When ), its value is 0. Kronecker The function ensures that only node pairs belonging to the same community will affect the modularity. This contributes to the overall brain connectivity analysis. The final output is a brain functional connectivity map with connectivity strength indicators, presented in a two-dimensional or three-dimensional visualization. Node color and size represent the level of connectivity strength, and edge thickness represents the level of connectivity. The values ​​are accompanied by community boundary markings, making it easy to intuitively analyze whole-brain functional connectivity patterns.

[0084] Stimulation parameter optimization stage: Establish the spatiotemporal correlation between the whole brain electrical activity distribution map and the brain functional connectivity map, align the stimulation parameters using a multi-objective optimization algorithm, and integrate the whole brain electrical activity distribution map and the brain functional connectivity map through a fusion network to generate an optimized stimulation parameter set.

[0085] The specific implementation process is as follows: First, establish the spatiotemporal correlation between the whole-brain electrical activity distribution map and the brain functional connectivity map. A global coordinate system is established using a preset reference point of the virtual brain model (usually defined as the midpoint between the anterior and posterior commissures). This coordinate system adopts a three-dimensional Cartesian coordinate system, with its origin located at the reference point, the X-axis pointing towards the right ear, the Y-axis towards the root of the nose, and the Z-axis towards the top of the head. The coordinate units are uniformly set to millimeters. The electrode coordinates in the whole-brain electrical activity distribution map are mapped to the global coordinate system through an affine transformation. The specific transformation formula is as follows: ;in, This represents the coordinates of the electrode in the local coordinate system; It is a rotation matrix; It is a translation vector; These are the mapped global coordinates. Through the above mapping, EEG activity distribution data in the global coordinate system is obtained. The data format is a multidimensional array containing the spatial coordinates of each electrode. The data includes the corresponding EEG activity intensity values. Simultaneously, the coordinates of connection communities in the brain functional connectivity map are mapped to the global coordinate system using the same transformation rules to obtain functional connectivity data in the global coordinate system. This data includes the center coordinates of each connection community, the connection intensity values ​​of nodes within the community, and the weights of connection edges between communities. In the global coordinate system, the EEG activity distribution data and functional connectivity data are spatially registered. The nearest neighbor interpolation algorithm is used to resample the EEG activity intensity values ​​to the connection community coordinate grid to ensure consistent spatial resolution. The registration accuracy is verified by calculating spatial overlap indices (such as the Dice coefficient). A Dice coefficient higher than 0.85 indicates valid spatial association. Based on a unified time reference, the timestamps of the EEG activity distribution data and functional connectivity data are aligned. The time reference is set to the initial acquisition time of the simulated EEG signal data as zero. Linear interpolation is used to synchronize the time series of the two types of data to the same time axis, with a sampling interval of 10 milliseconds. Temporal correlation analysis (such as cross-correlation function) is used to verify the consistency of temporal association. A maximum correlation coefficient exceeding 0.9 confirms successful establishment of temporal association.

[0086] Subsequently, a multi-objective optimization algorithm was used to align the stimulus parameters, and a fusion network was used to integrate the whole-brain electroencephalogram (EEG) distribution map and brain functional connectivity map to generate an optimized stimulus parameter set. The fusion network adopted a deep learning architecture, including a data input layer, a feature transformation layer, a cross-modal interaction layer, and a confidence output layer. The data input layer received EEG distribution data and functional connectivity data in a global coordinate system. The EEG distribution data was input in the form of a three-dimensional tensor, and the functional connectivity data was input in the form of an adjacency matrix. In the feature transformation layer, the EEG distribution data was processed by a three-dimensional convolutional network (with convolution kernel size...). Feature extraction is performed using a step size of 1 and a padding method of "same padding"), outputting an active feature vector with dimensions of [missing information]. (Typically set to 256 dimensions); Functional connectivity data is achieved through a graph convolutional network (based on spectral graph convolution theory, convolution kernel size...). Encoding with ReLU activation function, the output is a connection feature vector with dimension 1. (Typically set to 256 dimensions). Through a cross-modal interaction layer, feature cross-fusion is performed on the active feature vector and the connection feature vector. This layer employs a multi-head attention mechanism to calculate the mutual information weights between the active and connection features, generating a fused feature vector. The formula is as follows: ;in, For querying, key-value projection of activity feature vectors; For query, key, and value projections that connect feature vectors; For scaling dot product attention functions, This represents a vector concatenation operation. The dimension of the fused feature vectors is... (Typically set to 512 dimensions). The fused feature vector is input to the confidence output layer, which is composed of a fully connected network (128 hidden layers, sigmoid activation function), to calculate the initial stimulus parameter set, including stimulus frequencies. Stimulation intensity and stimulation location .

[0087] Using the initial stimulus parameter set as input, a multi-objective optimization algorithm (NSGA-II, a non-dominated sorting genetic algorithm, is invoked in this embodiment). The optimization objectives include connectivity enhancement, state stability, and resource economy. The connectivity enhancement objective aims to maximize the phase synchronization strength of critical connectivity communities, and its mathematical expression is: ;in, The score represents the connectivity enhancement objective; a higher value indicates a stronger phase synchronization in the key communities and a better connectivity enhancement effect. Indicates the number of key connected communities; For the community The average phase lock value. The state stability objective aims to minimize the spatiotemporal fluctuations of EEG activity, and its formula is: ;in, The score represents the goal of state stability. Since it is negative, the larger the value, the smaller the fluctuation of brain electrical activity and connection strength, and the more stable the state. Indicates the length of the time window; Indicates a point in time The standard deviation of brain electrical activity intensity; Indicates a point in time The standard deviation of connection strength. The resource-economic objective aims to minimize stimulus energy consumption, and its formula is: ;in, The score represents the resource economic objective. Since it is negative, the larger the value, the less total energy consumption of the stimulus parameter and the better the resource economy. and These are the stimulation frequency parameter and the stimulation intensity parameter, respectively. It represents the sum of squares of all stimulus frequency parameters, reflecting the total energy consumption of the frequency; This represents the sum of squares of all stimulus intensity parameters, reflecting the total energy consumption of the intensity. Constraints include physically immutable constraints (e.g., stimulus frequency range limited to 1-100 Hz, stimulus intensity range limited to 0.1-5.0 μA, stimulus location must be within the effective region of the virtual brain model) and policy adjustable constraints (e.g., total stimulus duration not exceeding 30 minutes, single stimulus duration not exceeding 10 seconds). Under these constraints, a multi-objective optimization algorithm performs parallel evaluations on multiple randomly generated stimulus policy parameter schemes (population size set to 100), calculating the achievement score of each stimulus policy parameter scheme relative to each optimization objective. The achievement score is normalized and mapped to the [0,1] interval. Based on preset weights, the achievement scores are weighted and aggregated to obtain the corresponding comprehensive evaluation value of the scheme. The calculation formula is as follows: ;in, This represents the overall evaluation value of the scheme. The higher the value, the better the overall performance of the stimulus strategy in terms of connectivity enhancement, state stability, and resource economy. , , These represent the weights for connection enhancement, state stability, and resource economy objectives, respectively, with preset weight values ​​of [values ​​to be filled in]. , , The stimulus strategy parameter scheme with the highest comprehensive evaluation value is selected from all stimulus strategy parameter schemes. The stimulus frequency, stimulus intensity, and stimulus location parameters contained therein are combined and encoded. The stimulus frequency and intensity parameters are encoded using floating-point numbers, and the stimulus location parameters are encoded using integer indices (corresponding to virtual electrode addresses). The final output is an optimized stimulus parameter set, which is stored in a structured data format (such as JSON) including an array of stimulus frequencies, an array of stimulus intensities, and a list of stimulus location coordinates.

[0088] Executable stimulus instruction conversion stage: Based on the adaptive control model, the optimized stimulus parameter set is converted into executable stimulus instructions.

[0089] The specific implementation process is as follows: First, the optimized stimulus parameter set is input into the adaptive control model in a structured data format (such as JSON). This dataset contains an array of stimulus frequencies, an array of stimulus intensities, and a list of stimulus location coordinates. The adaptive control model parses these parameters according to a preset instruction mapping protocol: the instruction mapping protocol is a rule base that defines the conversion logic from parameters to instructions, including data type validation, range validation, and encoding rules. For the stimulus frequency parameter, the model converts it into periodic timing trigger instructions. The timing trigger instructions are generated by the digital signal processing module, which maps frequency values ​​to pulse intervals, such as stimulus frequency... (Unit: Hz) Corresponding period Through formula Calculations show that, in seconds, a periodic square wave pulse sequence is generated using a high-precision timer (such as an FPGA-based clock divider), with each rising edge of the pulse marking the start of a stimulus event. Timing trigger commands are output as a binary code stream and managed through a buffer queue to avoid timing jitter. Simultaneously, the model integrates a phase-locked loop mechanism to ensure the pulse sequence is synchronized with the global clock, preventing frequency drift from affecting stimulus stability.

[0090] For the stimulus intensity parameter, the adaptive control model quantizes it into an output amplitude command of analog voltage or current. The generation of the output amplitude command relies on a digital-to-analog converter module, which converts the digitized intensity value into a continuous analog signal. Specifically, the stimulus intensity parameter... (Unit: microamps) After normalization, this is mapped to the input dynamic range of the digital-to-analog converter (DAC). For example, if the DAC resolution is 12 bits and the reference voltage is 5V, the formula for calculating the output amplitude command is: ;in, The input code represents the digital-to-analog converter; The maximum stimulus intensity (e.g., 5.0 μA); Indicates the normalization strength; The resolution is determined by the digital-to-analog converter (DAC), where 12 indicates that it is a 12-bit DAC. The signal is sent to a digital-to-analog converter via a serial peripheral interface to generate a corresponding analog voltage signal. This signal is then adjusted by a gain-adjustable amplifier (such as a programmable gate amplifier, PGA) to ensure the stimulation intensity remains within a safe range (e.g., 0.1-5.0 μA) and is linearly controllable. The model also monitors the stability of the output amplitude in real time, calibrating deviations through the analog-to-digital converter feedback loop to prevent overshoot or undershoot. For stimulation position parameters, the adaptive control model converts them into address gating instructions for the target virtual electrode using a spatial encoder. The spatial encoder maps the stimulation position coordinates to specific virtual electrode addresses based on the global coordinate system of the virtual brain model (a three-dimensional Cartesian system in millimeters). The address gating instructions use multiplexing technology, sending gating signals to the target electrode via the address bus. Each virtual electrode has a unique address code (e.g., a 16-bit integer). The spatial encoder converts the coordinates into address codes using a lookup table or hash function and generates gating instructions (e.g., a high-level active chip select signal). The gating command is output through the digital input / output module, activating the stimulation receiving circuit of the target virtual electrode. Simultaneously, the model verifies the address validity, ensuring the gating electrode is located within the effective region of the virtual brain model (e.g., avoiding gating the background area or invalid coordinates). The adaptive control model also integrates a real-time monitoring and adaptive adjustment mechanism. This mechanism compares the expected stimulus response (predicted based on whole-brain activity distribution maps and brain connectivity maps) with the actual response (collected from the virtual brain model) by simulating feedback from EEG signal data. If the deviation exceeds a preset threshold (e.g., the mean square error of EEG activity intensity is greater than 0.1), the model dynamically adjusts the command mapping parameters, for example, by fine-tuning the period of the stimulation frequency or the gain coefficient of the stimulation intensity using a proportional-integral-derivative controller to achieve closed-loop optimization. The entire stimulation modulation simulation phase is executed on a software-in-the-loop simulation platform. The adaptive control model, as the core controller, interacts with the virtual brain model through an application programming interface (API), generating executable stimulation commands and simulating the neural response after stimulation, thus providing a reliable and efficient testing environment for personalized modulation of consciousness disorders.

[0091] This embodiment also discloses a consciousness disorder stimulation and modulation system that integrates brainwave connectivity recognition, including the following modules:

[0092] The simulated EEG signal data acquisition module is configured to execute the steps of the simulated EEG signal data acquisition phase.

[0093] The connection identification and analysis module is configured to perform the steps of the connection identification and analysis phase.

[0094] The stimulus parameter optimization module is configured to execute the steps of the stimulus parameter optimization phase.

[0095] An executable stimulus instruction conversion module is configured to execute the steps of the stimulus regulation simulation phase.

[0096] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for regulating consciousness disorders by integrating brainwave connectivity recognition, characterized in that, Includes the following steps: Simulated EEG signal data acquisition stage: Simulated EEG signal data is generated based on the EEG signal generation model. The EEG signal generation model is a computational model that integrates the functional region structure defined by the standard brain functional atlas with the neuronal population dynamics described by the neuronal mass differential equation. Spatiotemporal filtering is applied to the simulated EEG signal data to generate a whole brain electrical activity distribution map, and key connectivity regions are identified in the whole brain electrical activity distribution map. Connectivity identification and analysis phase: Based on the identified key connectivity regions, a functional connectivity analysis network is deployed, and the connectivity strength between simulated EEG signal data is analyzed using a phase synchronization algorithm to generate a brain functional connectivity map; Stimulation parameter optimization stage: Establish the spatiotemporal correlation between the whole brain electrical activity distribution map and the brain functional connectivity map, align the stimulation parameters using a multi-objective optimization algorithm, and integrate the whole brain electrical activity distribution map and the brain functional connectivity map through a fusion network to generate an optimized stimulation parameter set; include: A global coordinate system is established using the preset reference points of the virtual brain model; The electrode coordinates in the whole brain activity distribution map are mapped to the global coordinate system to obtain brain activity distribution data in the global coordinate system. The fusion network comprises a data input layer, a feature transformation layer, a cross-modal interaction layer, and a confidence output layer. EEG activity distribution data and functional connectivity data in the global coordinate system are input to the data input layer. In the feature transformation layer, the EEG activity distribution data is converted into activity feature vectors, and the functional connectivity data is converted into connectivity feature vectors. Through the cross-modal interaction layer, the activity feature vectors and connectivity feature vectors are cross-fused to generate a fused feature vector. The fused feature vector is input to the confidence output layer to calculate the initial stimulus parameter set. Using the initial set of stimulation parameters as input, a multi-objective optimization algorithm is invoked, including a connection enhancement objective aimed at maximizing the phase synchronization strength of key connection communities, a state stability objective aimed at minimizing the spatiotemporal fluctuations of EEG activity, and a resource economy objective aimed at minimizing stimulation energy consumption as optimization objectives, and constrained conditions including physically immutable constraints and policy adjustable constraints are configured. Under the constraints, the multi-objective optimization algorithm evaluates multiple randomly generated stimulus strategy parameter schemes in parallel, calculates the achievement score of each stimulus strategy parameter scheme relative to each optimization objective, and aggregates the achievement scores based on preset weights to obtain the corresponding comprehensive evaluation value of the scheme. The stimulus strategy parameter scheme with the highest comprehensive evaluation value is selected from all stimulus strategy parameter schemes. The stimulus frequency, stimulus intensity and stimulus location parameters contained therein are combined and encoded, and the final output is the optimized stimulus parameter set. Executable stimulus instruction conversion stage: Based on the adaptive control model, the optimized stimulus parameter set is converted into executable stimulus instructions.

2. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 1, characterized in that, During the simulated EEG signal data acquisition phase: The standard brain functional atlas predefines key functional network regions, including the default mode network region, the executive control network region, and the salience network region; the neuronal quality differential equation generates EEG waveforms containing different frequency bands by simulating the average membrane potential and synaptic transmission dynamics of a neuronal population.

3. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 2, characterized in that, In the simulated EEG signal data acquisition stage, the steps for generating simulated EEG signal data include: Based on the standard brain functional atlas, the default mode network region, executive control network region, and salience network region are initialized as high-sensitivity focusing regions, and the remaining brain regions are initialized as low-sensitivity background regions; and a sensitivity function is constructed to divide the EEG monitoring space into high-sensitivity focusing regions and low-sensitivity background regions; The simulated EEG signal data is analyzed in real time through a preset anomaly response network. The input to the anomaly response network is the simulated EEG signal data, and the output is transient feature activity areas. The focusing area is dynamically updated based on these transient feature activity areas. If a new transient feature activity area is detected, it is upgraded to a high-sensitivity focusing area. If no transient feature activity is detected in an existing high-sensitivity focusing area, it is downgraded to a low-sensitivity background area. The sampling frequency of the high-sensitivity focusing area is higher than that of the low-sensitivity background area. The simulated EEG signal data is obtained by dynamically adjusting the virtual electrode operating parameters through configuration commands.

4. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 3, characterized in that, In the simulated EEG signal data acquisition stage, the step of applying spatiotemporal filtering to the simulated EEG signal data to generate a whole-brain electrical activity distribution map includes: Bandpass filtering and normalization are performed on the simulated EEG signal data, and the normalized simulated EEG signal data is integrated into a multidimensional tensor according to the virtual electrode channel dimension; A spatiotemporal coding network architecture is constructed, which includes a forward feature extraction path, a reverse feature reconstruction path, and a cross-scale feature fusion path. The integrated multidimensional tensor is input into the spatiotemporal coding network architecture; in the forward feature extraction path, a deep convolutional network is used to compress and abstract the features of the multidimensional tensor to generate a multi-scale feature mapping set, which includes micro-transient features, meso-mode features and macro-trend features. In the cross-scale feature fusion path, the micro-transient features and meso-mode features obtained in the forward feature extraction path are respectively passed to the corresponding resolution levels in the reverse feature reconstruction path. In the reverse feature reconstruction path, the macro trend features are upsampled to the meso scale through bilinear interpolation and fused with the meso mode features from the cross-scale feature fusion path; then the fused meso features are restored to the micro scale through interpolation and fused with the micro transient features from the cross-scale feature fusion path, finally outputting a high-resolution feature map that combines detail and wholeness. The high-resolution feature maps output by the reverse feature reconstruction path are superimposed and reconstructed to generate the whole brain electrical activity distribution map.

5. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 4, characterized in that, In the simulated EEG signal data acquisition phase, the step of identifying key connectivity regions in the whole-brain electrical activity distribution map includes: Based on the whole brain activity distribution map, the connection strength score of each virtual electrode channel is calculated. The connection strength score is obtained by extracting the intensity of the micro transient features, the stability of the meso mode features and the consistency of the macro trend features fused at each spatial location point in the whole brain activity distribution map, and performing nonlinear weighted aggregation. A primary threshold and a secondary threshold for connection strength scoring are preset. The connection strength score is compared with both the primary and secondary thresholds. If the connection strength score is lower than the primary threshold, the corresponding channel is marked as a low connection domain. If the connection strength score is higher than the primary threshold but lower than the secondary threshold, the corresponding channel is marked as a medium connection domain. If the connection strength score is higher than the secondary threshold, the corresponding channel is marked as a high connection domain. Different color codes were used to identify key connectivity regions in the whole brain activity distribution map. Different colors corresponded to different connectivity strength levels. Low connectivity regions were defined as weak connectivity regions, medium connectivity regions as medium connectivity regions, and high connectivity regions as strong connectivity regions.

6. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 1, characterized in that, The steps in the connection identification and analysis phase include: Based on the identified key connectivity regions, a high-density electrode simulation array is deployed at the corresponding location in the virtual brain model; Configure the analysis parameters of the functional connectivity analysis network and implement differentiated connectivity analysis strategies for different connectivity regions: configure a first analysis window for high connectivity regions, a second analysis window for medium connectivity regions, and a third analysis window for low connectivity regions; wherein the duration of the first analysis window is shorter than the duration of the second analysis window, and the duration of the second analysis window is shorter than the duration of the third analysis window. Through the aforementioned functional connectivity analysis network, a phase synchronization algorithm is applied to perform functional connectivity analysis on the simulated EEG signal data collected by the high-density electrode simulation array. A multi-band coupled phase correction model is introduced, and the synchronization parameters in the phase correction model are continuously optimized through an iterative phase analysis algorithm. The iteration is terminated when the change in the phase consistency index is lower than the preset convergence threshold. A brain functional connectivity matrix is ​​constructed based on graph theory network analysis algorithms. Key connectivity communities are calculated using community detection algorithms, and finally, a brain functional connectivity map with connectivity strength labels is output.

7. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 6, characterized in that, In the stimulation parameter optimization phase, the step of establishing the spatiotemporal correlation between the whole-brain electrical activity distribution map and the brain functional connectivity map includes: Map the coordinates of the connectivity communities in the brain functional connectivity atlas to the global coordinate system to obtain functional connectivity data in the global coordinate system; In the global coordinate system, the EEG activity distribution data and the functional connectivity data are spatially registered to establish a spatial association; Based on a unified time reference, the timestamps of the EEG activity distribution data and the functional connectivity data are aligned to establish a time correlation.

8. The method for regulating consciousness disorders by integrating EEG connectivity recognition according to claim 1, characterized in that, The steps of the executable stimulus instruction switching phase include: The optimized stimulus parameter set is input into the adaptive control model. The adaptive control model compiles the stimulus frequency, stimulus intensity, and stimulus position parameters in the optimized stimulus parameter set into executable stimulus instructions according to a preset instruction mapping protocol. Among them, the stimulus frequency parameter is converted into a periodic timing trigger instruction, the stimulus intensity parameter is quantized into an output amplitude instruction of analog voltage or current, and the stimulus position parameter is converted into an address gating instruction of the target virtual electrode through a spatial encoder.

9. A consciousness disorder stimulation modulation system integrating EEG connectivity recognition, applied to the consciousness disorder stimulation modulation method integrating EEG connectivity recognition as described in any one of claims 1-8, characterized in that, It includes the following modules: The simulated EEG signal data acquisition module is configured to execute the steps of the simulated EEG signal data acquisition phase. The connection identification and analysis module is configured to perform the steps of the connection identification and analysis phase. The stimulus parameter optimization module is configured to execute the steps of the stimulus parameter optimization phase. An executable stimulus instruction conversion module is configured to execute the steps of the stimulus regulation simulation phase.

Images