A neural regulation target positioning system and method based on spatiotemporal cascade evolution features

Abstract

The application discloses a neural regulation target positioning system and method based on space-time cascade evolution characteristics, comprising the following steps: acquiring synchronous transcranial magnetic stimulation-electroencephalogram offline data; preprocessing the data and extracting scalp potential signals, mapping to the cortical source space through a source reconstruction algorithm, and obtaining a source time sequence; constructing a time-varying cross-frequency directed brain network based on an adaptive direct transfer function, slidingly calculating the whole brain directed connection strength in multiple time windows, and extracting a set of space-time cascade evolution characteristics of early sensory gating, mid-term processing and late network disinhibition; screening the optimal characteristic value, and outputting the corresponding brain area three-dimensional coordinates and reference regulation parameters. The application realizes the full-process automation from signal processing to target coordinate generation, improves the spatial resolution, analyzes the neural network dynamic evolution mechanism, and provides precise positioning support for neural regulation research.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interface and artificial intelligence technology, and particularly relates to a neural modulation target localization system and method based on spatiotemporal cascade evolution characteristics. Background Technology

[0002] In the field of neuromodulation technology research, the accuracy of target localization is a key factor affecting the effectiveness of modulation research. Existing neuroimaging data processing and localization techniques mainly suffer from the following limitations: Insufficient spatial resolution: Although traditional scalp EEG technology has high temporal resolution, the potential signals recorded by the scalp are severely blurred due to the low conductivity of the skull and the volume conduction effect of cerebrospinal fluid. It is difficult to accurately locate the signal source in the deep cortex or medial functional area, resulting in a large spatial error in the localization results.

[0003] Lack of dynamic mechanism analysis: While existing techniques such as functional magnetic resonance imaging (fMRI) offer high spatial resolution, their temporal resolution is insufficient, failing to capture millisecond-level dynamic interaction information of neural networks. Existing static network analysis methods struggle to analyze the dynamic mechanisms evolving over time during signal transmission, thus failing to meet the need for precise localization of transient neural activity.

[0004] Feature extraction is limited to a single dimension: Existing technologies often focus on features at a single time point or in a single frequency band, ignoring the cascading evolution of neural signals over time. Furthermore, they lack a panoramic quantification and screening mechanism for the strength of connections across the entire brain network, making the localization results susceptible to local noise interference and lacking overall optimality.

[0005] Source reconstruction technology, by constructing a head model and combining it with signal processing algorithms, can inversely map low-resolution signals recorded on the scalp to the cortical source space, providing a technical foundation for improving the spatial accuracy of target localization. Therefore, this invention proposes a localization system and method based on spatiotemporal cascaded evolution features, aiming to solve the technical challenges of existing technologies in dynamic feature extraction, whole-brain network quantization, and accurate target localization. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a neural modulation target localization system and method based on spatiotemporal cascade evolution characteristics, thereby resolving the issues present in the prior art.

[0007] To achieve the above objectives, this invention provides a neural modulation target localization system based on spatiotemporal cascade evolution characteristics, comprising: The data input module is used to acquire pre-stored synchronous transcranial magnetic stimulation-electroencephalography offline data; The source reconstruction processing module is used to preprocess the synchronous transcranial magnetic stimulation-electroencephalography offline data and extract scalp potential signals. The scalp potential signals are then mapped to the cortical source space in reverse using the source reconstruction algorithm to obtain the source time series of each brain region of interest. The spatiotemporal network analysis module is used to construct a time-varying cross-frequency directed brain network based on the source time series through an adaptive directional transfer function, and to calculate the directed connection strength of the whole brain regions based on the time-varying cross-frequency directed brain network within multiple time windows after stimulation, and to extract the spatiotemporal cascade evolution feature set. The target localization output module is used to filter the optimal feature value from the spatiotemporal cascaded evolution feature set. When the optimal feature value exceeds the preset threshold, it calculates and outputs the three-dimensional spatial coordinates of the corresponding brain region node and the reference control parameters corresponding to the brain region.

[0008] Optionally, the source reconstruction processing module includes a preprocessing unit and a source localization unit; The preprocessing unit assigns scalp coordinates to each electrode, removes invalid electrode channels, extracts EEG data for preset time periods before and after stimulation, performs baseline correction using a baseline window before stimulation, deletes data segments within a set time period before and after stimulation pulses, removes noise artifacts through two independent component analyses, performs interpolation repair on bad channels, uses bandpass filtering to eliminate noise, and uses the average reference as the heavy reference to obtain the preprocessed scalp signal. The source localization unit uses a standardized low-resolution EEG algorithm to project the preprocessed scalp signal onto multiple brain regions defined by a standard brain atlas, and calculates the average current density of each brain region as the source time series.

[0009] Optionally, the spatiotemporal network analysis module includes a cross-frequency network construction unit and a feature extraction unit; The cross-frequency network construction unit is used to obtain the directional connection matrix of the time-varying cross-frequency directed brain network based on the source time series; The feature extraction unit is used to calculate the directed connectivity strength of whole-brain regions within multiple time windows after stimulation based on the directed connectivity matrix, and extract the spatiotemporal cascade evolution feature set.

[0010] Optionally, the cross-frequency network construction unit is used to establish a multivariate adaptive autoregressive model based on the source time series, determine the model order through the Akaike information criterion, solve the time-varying coefficient matrix using the multivariate adaptive autoregressive model and the model order, perform a frequency domain transformation on the time-varying coefficient matrix to obtain a frequency domain transfer function matrix, calculate the normalized adaptive directional transfer function value based on the frequency domain transfer function matrix, and divide it according to multiple preset frequency bands to obtain the frequency band directional connection matrix of the time-varying cross-frequency directed brain network; wherein, the normalized adaptive directional transfer function value quantifies the directional connection strength between nodes in different frequency bands between 0 and 1, and the preset frequency bands include δ, θ, α, and β frequency bands.

[0011] Optionally, the feature extraction unit is used to calculate the directed connection strength between multiple brain regions within a preset time range after stimulation, based on the directional connection matrix of the time-varying cross-frequency directed brain network, by setting a step size sliding time window, and extracting the early sensory gating feature set in the early time window, the mid-processing dynamic network feature set in the mid-time window, and the late network desuppression feature set in the late time window, to obtain the spatiotemporal cascaded evolution feature set.

[0012] Optionally, the target localization output module includes a feature filtering unit and a coordinate calculation unit; The feature selection unit is used to select the optimal feature value from the spatiotemporal cascaded evolution feature set using a recursive feature elimination method; The coordinate calculation unit is used to retrieve the three-dimensional coordinates of the brain region associated with the optimal feature value in the standard brain map according to the comparison attribute corresponding to the optimal feature value, calculate the centroid coordinates of the associated brain region as the target coordinates, and output the target coordinates and the corresponding reference control parameters.

[0013] Optionally, the offline data is recorded by an EEG system equipped with multiple electrodes arranged in a standard layout, and the data format includes time-stamped information of the triggering stimulus.

[0014] This invention also provides a method for locating neural modulation targets based on spatiotemporal cascade evolution characteristics, applied to the above-mentioned system, comprising: Acquire pre-stored synchronous transcranial magnetic stimulation-electroencephalography offline data; The synchronized transcranial magnetic stimulation-electroencephalography offline data were preprocessed to obtain scalp potential signals; The scalp potential signals are inversely mapped to the cortical source space using a source reconstruction algorithm to obtain the source time series of each brain region of interest. Based on the source time series, a time-varying cross-frequency directed brain network is constructed through an adaptive directional transfer function. Within multiple time windows after stimulation, the directed connection strength of the whole brain regions is calculated by sliding based on the time-varying cross-frequency directed brain network, and the spatiotemporal cascade evolution feature set is extracted. The optimal feature value is selected from the spatiotemporal cascaded evolution feature set. When the optimal feature value exceeds the preset threshold, the corresponding three-dimensional spatial coordinates of the brain region node and the reference control parameters are calculated and output.

[0015] Optionally, constructing the time-varying cross-frequency directed brain network includes: A multivariate adaptive autoregressive model is established based on the source time series. The model order is determined by the Akaike information criterion. The time-varying coefficient matrix is ​​solved using the model and the model order. The time-varying coefficient matrix is ​​transformed in the frequency domain to obtain the frequency domain transfer function matrix; The normalized adaptive directional transfer function value is calculated based on the frequency domain transfer function matrix, and divided into multiple preset frequency bands to obtain the frequency band directional connection matrix of the time-varying cross-frequency directed brain network; wherein, the normalized adaptive directional transfer function value quantifies the directional connection strength between nodes in different frequency bands between 0 and 1, and the preset frequency bands include δ, θ, α, and β frequency bands.

[0016] Optionally, extracting the spatiotemporal cascaded evolution feature set includes: Based on the directional connectivity matrix of the time-varying cross-frequency directed brain network, a step-size sliding time window is set to calculate the directional connectivity strength between nodes in multiple brain regions within a preset time range after stimulation; the whole-brain connectivity matrix within the early time window is locked as the early sensory gating feature set; the whole-brain connectivity matrix within the middle time window is locked as the middle processing dynamic network feature set; and the whole-brain connectivity matrix within the late time window is locked as the late network de-inhibition feature set, thus obtaining the spatiotemporal cascade evolution feature set.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: (1) Spatiotemporal dual-dimensional high-precision imaging: Combining the millisecond-level temporal resolution of synchronous transcranial magnetic stimulation-electroencephalography (TMS-EEG) with the brain region-level spatial resolution of source reconstruction technology, the neural activity changes of deep cognitive nuclei during dynamic cognitive processing were successfully captured, solving the technical problem of locating deep brain regions by traditional methods.

[0018] (2) Depth of mechanism analysis: The cross-frequency ADTF algorithm and the full-time sliding window calculation mechanism were introduced to capture the spatiotemporal cascade evolution characteristics from early sensory gating, mid-term processing to late-term network regulation within the range of 0-300ms, making the localization results more consistent with the dynamic evolution law of brain neural activity.

[0019] (3) Global optimization localization: By calculating the functional connectivity strength between nodes in multiple brain regions of the whole brain and introducing a feature screening mechanism, the interference of redundant features is eliminated, and the most discriminative optimal feature value is determined at the level of the whole brain network. This avoids the one-sidedness of local feature calculation and realizes accurate coordinate calculation based on individualized brain network functional state. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a structural diagram of a neural modulation target localization system based on spatiotemporal cascade evolution characteristics, according to an embodiment of the present invention. Figure 2 This is a flowchart of a neural modulation target localization method based on spatiotemporal cascade evolution characteristics, according to an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Example 1 like Figure 1 As shown, this embodiment provides a neural modulation target localization system based on spatiotemporal cascade evolution characteristics, including a data input module, a source reconstruction processing module, a spatiotemporal network analysis module, and a target localization output module.

[0024] The specific architecture is as follows: 1. Data Input Module: This module is configured with a standard data interface for acquiring pre-stored TMS-EEG offline data. The offline data is a digital EEG time-series file conforming to internationally accepted formats, containing time-stamped information of trigger stimuli, and originating from public databases, previous clinical research data, or simulation-generated data.

[0025] The offline data in the data input module, and the corresponding EEG signals, were recorded by the ANT Neuro EEG system, which is equipped with 64 electrodes arranged in the international 10-20 layout.

[0026] 2. Source Reconstruction Processing Module: Configured to perform standardized preprocessing on the offline data, construct an individualized head model, and use a source reconstruction algorithm to inversely map scalp potential signals to the cortical source space to obtain high temporal resolution source time series for each brain region of interest; this module includes a preprocessing unit and a source localization unit; (1) Preprocessing unit: configured to perform a high-fidelity signal cleaning process. Specifically, it includes: data segmentation extraction based on stimulus labels; adaptive separation and removal of electromyography, electrooculography and power frequency noise artifacts using independent component analysis (ICA) technology; high-order bandpass filtering and notch filtering to eliminate baseline drift and power frequency interference; and finally, average reference reset to output scalp potential data with high signal-to-noise ratio.

[0027] Specifically, a high-fidelity signal cleaning process was performed on the raw TMS-EEG offline data: precise scalp coordinates were assigned to each electrode, invalid electrodes were removed, time history data from -1000ms to 1000ms was extracted, and correction was performed using -500ms to -20ms as the baseline; strong artifact data segments from -2ms to 10ms before and after the TMS pulse were deleted, and EMG, eye movement, and electrode noise were removed sequentially through two rounds of independent component analysis (ICA), and bad channels were repaired by interpolation; 1-45Hz bandpass filtering and 50 Hz notch filtering were used to eliminate noise interference, and the average reference was used as the primary reference.

[0028] (2) Source localization unit: configured to construct an individualized head model, using the sLORETA algorithm to inversely map scalp potential signals to the cortical source space. The core calculation formula is: (1) (2) (3) In the formula, This is a measurement of scalp potential. For the lead field matrix, For current source distribution, It is Gaussian white noise. It is a linear inverse operator. The source covariance normalized matrix, For the first The standardized current density values ​​of the individual units. The unit projects the signal onto 68 brain regions of interest as defined by the standard Desikan-Killiany atlas, and calculates the average current density of each brain region as the source time series (source spatial signal).

[0029] 3. Spatiotemporal Network Analysis Module: This module is the core computing engine of the system. Based on the source time series, it constructs a time-varying cross-frequency directed brain network through the Adaptive Directed Transfer Function (ADTF), quantifies the information flow and connection strength between different frequency bands, and extracts the spatiotemporal cascade evolution feature set related to early sensory gating, mid-stage processing, and late-stage network regulation; including cross-frequency network construction unit and feature extraction unit; (1) Cross-frequency network building unit: configured to solve the cross-frequency adaptive directional transfer function (ADTF) based on a multivariate adaptive autoregressive model (MVAAR), the calculation formula is defined as: (4) In the formula, It is a multivariate time series vector. The time-varying coefficient matrix, For multivariate white noise, the optimal model order is... Determined by minimizing the Akaike Information Criterion (AIC): (5) It is the number of time series. For time points, This is the corresponding covariance matrix. Based on the obtained time-varying coefficient matrix... Performing a frequency domain transformation on it yields Finally, the directed information flow from j to i at each time point t is obtained: (6) (7) The normalized ADTF is defined in the range (0, 1) as: (8) (9) in, Ultimately, a directed connection is obtained for the entire frequency band; The cross-frequency network construction divides the ADTF results into frequency bands of δ (1-3 Hz), θ (4-7 Hz), α (8-13 Hz), and β (14-30 Hz).

[0030] (2) Feature Extraction Unit: Configured to perform sliding window computation across the entire brain. A time-frequency dynamic network is constructed with a sliding step size of 30ms. Within a time range of 0-300ms after stimulation, the directed connection strengths between nodes in 68 brain regions are calculated. Based on temporal evolution characteristics, the unit extracts three types of whole-brain feature sets: (a) Early sensory gating-related feature set: locking the whole-brain connectivity matrix within a 0-60ms time window; (b) Mid-processing dynamic network feature set: locking the whole brain connectivity matrix within a time window of 60-180ms, reflecting the transmission characteristics from the sensory cortex to the association cortex; (c) Late-stage network regulation-related feature set: locking the whole-brain connectivity matrix within a time window of 180-300ms, reflecting long-range connectivity and control characteristics.

[0031] The spatiotemporal network analysis module constructs a time-varying cross-frequency directed brain network based on source time series and extracts spatiotemporal cascade evolution features, breaking through the limitations of traditional single-band static network analysis. It quantifies the directional information flow between brain regions using the ADTF method, capturing the unique multi-stage evolutionary patterns of neural activity: the early sensory gating stage (0-60ms), the mid-processing stage (60-180ms), and the late network regulation stage (180-300ms), comprehensively analyzing the dynamic interaction mechanism of the neural network. This module employs a cross-frequency adaptive directional transfer function algorithm, calculating the directed connection strength between nodes in 68 brain regions across the entire brain within a 0-300ms time range with a 30ms step size, constructing a full-time dynamic network feature matrix.

[0032] 4. Target localization output module: The module uses a feature selection algorithm to determine the optimal feature value from the whole brain feature set. When the optimal feature value exceeds the preset threshold range, it calculates and outputs the three-dimensional spatial coordinates of the corresponding brain region node and the corresponding reference control parameters. This module includes a feature selection unit and a coordinate calculation unit. (1) Feature selection unit: configured to perform dimensionality reduction and optimization on the extracted early, middle and late whole brain feature sets. The recursive feature elimination method (RFE) or other feature importance evaluation algorithms are used to select the feature with the highest discriminative power as the optimal feature value in the whole brain feature dimension.

[0033] (2) Coordinate calculation unit: configured to execute localization logic based on the selected optimal feature: if the value of the optimal feature is lower or higher than the preset threshold, the system automatically locks the brain region node corresponding to the optimal feature, retrieves its three-dimensional spatial coordinates in the standard brain map, and outputs the coordinates and the corresponding reference regulation parameters as reference data related to the neural regulation target.

[0034] The target localization output module is configured to optimize and filter the extracted whole-brain feature set to obtain the optimal feature subset. When the optimal feature value exceeds a preset threshold range, it calculates and outputs the corresponding three-dimensional spatial coordinates of the brain region node and the corresponding reference regulation parameters. This module uses a feature filtering algorithm to determine the optimal feature subset from the whole-brain feature matrix, calculates the target coordinates based on the brain region node attributes corresponding to the optimal features, and uses the centroid coordinates of the target brain region as the target coordinates. The output reference regulation parameter data is determined based on the comparison between the optimal feature value and the preset threshold: if the comparison result is below the lower limit, the regulation reference parameter in the corresponding direction is output; if the comparison result is above the upper limit, the regulation reference parameter in the corresponding direction is output. The target coordinates and parameter data output by this system are data processing results and can be used as reference data for research related to neural regulation.

[0035] This system calculates the functional connectivity strength of the whole brain in 68×68 dimensions and introduces a feature selection mechanism to eliminate the interference of redundant features. It determines the most discriminative optimal feature value at the whole brain network level, avoids the one-sidedness of local feature calculation, and realizes accurate coordinate calculation based on individualized brain network functional state.

[0036] like Figure 2 As shown, this embodiment also provides a method for locating neural modulation targets based on spatiotemporal cascade evolution characteristics, including the following steps: S1. Offline data acquisition: Acquire pre-stored TMS-EEG offline data, which are derived from public databases, previous clinical research data, or simulation data generated by simulation. S2. Cortical source signal reconstruction: The offline data is standardized and preprocessed to construct an individualized head model. The source reconstruction algorithm is used to inversely map the scalp potential signal to the cortical source space to obtain high temporal resolution source time series of each brain region of interest. S3. Construction of Time-Varying Cross-Frequency Brain Network: Based on the source time series, a time-varying cross-frequency directed brain network is constructed through the adaptive directional transfer function (ADTF) to quantify the information flow direction and connection strength between different frequency bands; S4. Spatiotemporal cascade evolution feature extraction: Using a 30ms sliding window, the directed connection strength between 68 brain regions in the whole brain is calculated within the range of 0-300ms. Spatiotemporal cascade evolution feature sets (whole brain feature sets composed of whole brain connection matrices) related to early sensory gating, mid-processing and late network desuppression are extracted. S5. Target localization and parameter output: The optimal feature value is determined from the whole brain feature set using a feature selection algorithm. When the optimal feature value exceeds the preset threshold range, the corresponding three-dimensional spatial coordinates of the brain region node and the corresponding reference control parameters are calculated and output.

[0037] The specific process of constructing the time-varying cross-frequency directed brain network in step S3 is as follows: A multivariate adaptive autoregressive model was established, and the time-varying coefficient matrix was solved. The optimal model order was determined by minimizing the Akaike Information Criterion (AIC). Perform a frequency domain transformation on the time-varying coefficient matrix to obtain the frequency domain transfer function matrix; Calculate the normalized ADTF value to quantify the directional connection strength between nodes in different frequency bands; The ADTF results are divided by frequency band, and the data is divided into 30ms time windows. A time-frequency dynamic network is constructed to generate directional connection matrices for each frequency band and time window.

[0038] The feature selection in step S5 specifically involves using a recursive feature elimination method to select the features with the highest discriminative power from the whole-brain network connectivity features based on feature importance ranking, as the optimal features for the early, middle, and late stages.

[0039] Applications of neural modulation target localization: Based on the above system architecture and analysis process, using a simulated dataset as input, the system's target coordinates and control parameter output effects are displayed as follows: Within the early time window, when a certain feature value is lower than the preset threshold (e.g., the α-band connectivity value of the left fusiform gyrus-left entorhinal cortex is lower than the preset threshold), the system uses a recursive feature elimination method to select the connectivity as the optimal feature, locks the corresponding brain region node, and outputs the three-dimensional spatial coordinates of the left entorhinal cortex: X=-27.32mm, Y=-34.51mm, Z=-12.46mm. At the same time, it outputs the corresponding neural regulation reference parameters. These coordinates and parameters can be used as reference data for neural regulation-related research.

[0040] For features with values ​​exceeding a preset threshold within the late time window (e.g., the delta band connectivity value between the left superior temporal gyrus and the right inferior parietal lobule exceeds the preset threshold, or the beta band connectivity value between the left superior temporal gyrus and the right calcarine cortex exceeds the preset threshold), the system selects these connections as the optimal features, locks the corresponding brain region nodes, and outputs the three-dimensional spatial coordinates of the right inferior parietal lobule: X=52.18mm, Y=-45.36mm, Z=32.74mm. At the same time, it outputs the corresponding neural regulation reference parameters, which can be used as reference data for neural regulation-related research.

[0041] For the same simulated dataset, the system can repeatedly execute the full-process analysis at different time points and output the coordinates and parameter results of the same target point. The early sensory gating connection strength recovery rate and the late network connection strength decline rate are used as system stability evaluation indicators to verify the consistency and reliability of the system output, and provide data support for the longitudinal analysis of neuromodulation research.

[0042] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A neural modulation target localization system based on spatiotemporal cascade evolution characteristics, characterized in that, include: The data input module is used to acquire pre-stored synchronous transcranial magnetic stimulation-electroencephalography offline data; The source reconstruction processing module is used to preprocess the synchronous transcranial magnetic stimulation-electroencephalography offline data and extract scalp potential signals. The scalp potential signals are then mapped to the cortical source space in reverse using the source reconstruction algorithm to obtain the source time series of each brain region of interest. The spatiotemporal network analysis module is used to construct a time-varying cross-frequency directed brain network based on the source time series through an adaptive directional transfer function, and to calculate the directed connection strength of the whole brain regions based on the time-varying cross-frequency directed brain network within multiple time windows after stimulation, and to extract the spatiotemporal cascade evolution feature set. The spatiotemporal network analysis module includes a cross-frequency network construction unit and a feature extraction unit; The cross-frequency network construction unit is used to obtain the directional connection matrix of the time-varying cross-frequency directed brain network based on the source time series; The feature extraction unit is used to calculate the directed connectivity strength of whole-brain regions within multiple time windows after stimulation based on the directed connectivity matrix, and extract the spatiotemporal cascade evolution feature set. The feature extraction unit is used to calculate the directed connection strength between multiple brain regions within a preset time range after stimulation, based on the directional connection matrix of the time-varying cross-frequency directed brain network, by setting a step size sliding time window, and extracting the early sensory gating feature set in the early time window, the mid-processing dynamic network feature set in the mid-time window, and the late network desuppression feature set in the late time window, to obtain the spatiotemporal cascade evolution feature set. The target localization output module is used to filter the optimal feature value from the spatiotemporal cascaded evolution feature set. When the optimal feature value exceeds the preset threshold, it calculates and outputs the three-dimensional spatial coordinates of the corresponding brain region node and the reference control parameters corresponding to the brain region.

2. The neural modulation target localization system based on spatiotemporal cascade evolution characteristics according to claim 1, characterized in that, The source reconstruction processing module includes a preprocessing unit and a source localization unit; The preprocessing unit assigns scalp coordinates to each electrode, removes invalid electrode channels, extracts EEG data for preset time periods before and after stimulation, performs baseline correction using a baseline window before stimulation, deletes data segments within a set time period before and after stimulation pulses, removes noise artifacts through two independent component analyses, performs interpolation repair on bad channels, uses bandpass filtering to eliminate noise, and uses the average reference as the heavy reference to obtain the preprocessed scalp signal. The source localization unit uses a standardized low-resolution EEG algorithm to project the preprocessed scalp signal onto multiple brain regions defined by a standard brain atlas, and calculates the average current density of each brain region as the source time series.

3. The neural modulation target localization system based on spatiotemporal cascade evolution characteristics according to claim 1, characterized in that, The cross-frequency network construction unit is used to establish a multivariate adaptive autoregressive model based on the source time series, determine the model order using the Akaike information criterion, solve for the time-varying coefficient matrix using the multivariate adaptive autoregressive model and the model order, and perform a frequency domain transformation on the time-varying coefficient matrix to obtain a frequency domain transfer function matrix; calculate the normalized adaptive directional transfer function value based on the frequency domain transfer function matrix, and divide it according to multiple preset frequency bands to obtain the frequency band directional connection matrix of the time-varying cross-frequency directed brain network; wherein, the normalized adaptive directional transfer function value quantifies the directional connection strength between nodes in different frequency bands between 0 and 1, and the preset frequency bands include δ, θ, α, and β frequency bands.

4. The neural modulation target localization system based on spatiotemporal cascade evolution characteristics according to claim 1, characterized in that, The target localization output module includes a feature filtering unit and a coordinate calculation unit; The feature selection unit is used to select the optimal feature value from the spatiotemporal cascaded evolution feature set using a recursive feature elimination method; The coordinate calculation unit is used to retrieve the three-dimensional coordinates of the brain region associated with the optimal feature value in the standard brain map according to the comparison attribute corresponding to the optimal feature value, calculate the centroid coordinates of the associated brain region as the target coordinates, and output the target coordinates and the corresponding reference control parameters.

5. The neural modulation target localization system based on spatiotemporal cascade evolution characteristics according to claim 1, characterized in that, The offline data was recorded using an EEG system equipped with multiple electrodes arranged in a standard layout, and the data format included time-stamped information about the triggering stimuli.

6. A method for locating neural modulation targets based on spatiotemporal cascade evolution characteristics, applied to the system described in any one of claims 1-5, characterized in that, include: Acquire pre-stored synchronous transcranial magnetic stimulation-electroencephalography offline data; The synchronized transcranial magnetic stimulation-electroencephalography offline data were preprocessed to obtain scalp potential signals; The scalp potential signals are inversely mapped to the cortical source space using a source reconstruction algorithm to obtain the source time series of each brain region of interest. Based on the source time series, a time-varying cross-frequency directed brain network is constructed through an adaptive directional transfer function. Within multiple time windows after stimulation, the directed connection strength of the whole brain regions is calculated by sliding based on the time-varying cross-frequency directed brain network, and the spatiotemporal cascade evolution feature set is extracted. The optimal feature value is selected from the spatiotemporal cascaded evolution feature set. When the optimal feature value exceeds the preset threshold, the corresponding three-dimensional spatial coordinates of the brain region node and the reference control parameters are calculated and output.

7. The method for locating neural modulation targets based on spatiotemporal cascade evolution characteristics according to claim 6, characterized in that, Constructing the time-varying cross-frequency directed brain network includes: A multivariate adaptive autoregressive model is established based on the source time series. The model order is determined by the Akaike information criterion. The time-varying coefficient matrix is ​​solved using the model and the model order. The time-varying coefficient matrix is ​​transformed in the frequency domain to obtain the frequency domain transfer function matrix; The normalized adaptive directional transfer function value is calculated based on the frequency domain transfer function matrix, and divided into multiple preset frequency bands to obtain the frequency band directional connection matrix of the time-varying cross-frequency directed brain network; wherein, the normalized adaptive directional transfer function value quantifies the directional connection strength between nodes in different frequency bands between 0 and 1, and the preset frequency bands include δ, θ, α, and β frequency bands.

8. The method for locating neural modulation targets based on spatiotemporal cascade evolution characteristics according to claim 6, characterized in that, Extracting the spatiotemporal cascaded evolution feature set includes: Based on the directional connectivity matrix of the time-varying cross-frequency directed brain network, a step-size sliding time window is set to calculate the directional connectivity strength between nodes in multiple brain regions within a preset time range after stimulation; the whole-brain connectivity matrix within the early time window is locked as the early sensory gating feature set; the whole-brain connectivity matrix within the middle time window is locked as the middle processing dynamic network feature set; and the whole-brain connectivity matrix within the late time window is locked as the late network de-inhibition feature set, thus obtaining the spatiotemporal cascade evolution feature set.

Images