CN-120783950-B - Task state brain-computer interface training system for closed-loop transcranial magnetic stimulation

CN120783950BCN 120783950 BCN120783950 BCN 120783950BCN-120783950-B

Abstract

The invention discloses a task state brain-computer interface training system for closed-loop transcranial magnetic stimulation, which particularly relates to the field of cognitive rehabilitation assistance, and comprises the steps of synchronously collecting multichannel brain electrical signals and brain function activation records in the rehabilitation training process, dividing task segments, extracting frequency domain energy characteristics of each channel, and identifying state transition candidate segments based on frequency band energy distribution change; the system builds an activation state identifier by taking the topology vector and the brain function activation tag as supervision data, so as to realize the real-time judgment of different function states. The recognition result drives a stimulation parameter recommendation module, a preset stimulation strategy mapping table is queried according to the recognition state, and recommended stimulation parameters are dynamically generated to a magnetic stimulation execution interface, so that personalized closed-loop nerve regulation and control is realized.

Inventors

LIU RUI
JIANG JUN

Assignees

福建致远智创科技有限公司
福州脑机接口技术有限公司

Dates

Publication Date: 20260512
Application Date: 20250807

Claims (6)

1. The task state brain-computer interface training system for closed-loop transcranial magnetic stimulation is characterized by comprising a task segment dividing module, an energy feature extraction module, a state transition screening module, an activation topology construction module, a recognizer establishment module and a stimulation parameter recommendation module; the task segment dividing module collects brain electrical signals and synchronous brain function activation records during the execution of a rehabilitation training task, and constructs a task segment sequence; the energy characteristic extraction module performs frequency domain conversion on the electroencephalogram signals of different channels and extracts the energy characteristics of the electroencephalogram signals; the state transition screening module compares the frequency band energy distribution of the continuous task segments and screens state transition candidate segments; The activation topology construction module constructs a signal synchronization matrix based on time sequence brain signals of the state transition candidate segment, and analyzes task activation topology vectors crossing brain areas by using a graph convolution neural network; the identifier building module is used for building an identification classifier of the brain function activation state by taking the task activation topology vector and the brain function activation tag as training data; The stimulation parameter recommendation module outputs transcranial magnetic stimulation recommendation parameters in real time by identifying the current brain function activation state; the state transition screening module compares the frequency band energy distribution of the continuous task segments, and screens the state transition candidate segments specifically comprises: Dividing continuous time windows in a time sequence energy characteristic expression vector set of the electroencephalogram signal based on the boundary time of the task segment, and constructing an energy characteristic comparison candidate section index list; generating the frequency band energy distribution of the task segment level based on the two-dimensional energy matrix in each time window; selecting task segments corresponding to adjacent candidate segment indexes from the frequency band energy distribution set for comparison, and calculating a statistical difference measurement index between each pair of frequency band energy distribution; Screening task segment pairs with energy distribution structure difference measurement indexes exceeding a preset threshold value through threshold value comparison, marking corresponding task segments as energy mutation segments, and recording the start and stop time of the energy mutation segments as candidate segment boundaries; integrating the start and stop boundaries of the energy mutation fragments to form a state transition candidate segment; The active topology construction module constructs a signal synchronization matrix based on time sequence electroencephalogram signals of the state transition candidate section, and analyzes task active topology vectors crossing brain areas by using a graph convolution neural network, wherein the task active topology vectors specifically comprise: extracting an electroencephalogram signal sequence in a corresponding time period from a state transition candidate segment, and reconstructing a brain-crossing region time sequence signal array based on a channel index; calculating a phase locking value in the candidate segment between each pair of channel signals to generate a cross-channel signal correlation intensity matrix; converting the signal association intensity matrix into a weighted undirected graph structure, wherein graph nodes correspond to channel positions, and graph side weights reflect neural cooperative response intensities; Training and extracting features of the convolutional neural network model of each candidate segment graph structure input graph, and outputting high-order topological vectors of corresponding task segments; And organizing topology vectors corresponding to all state candidate segments in a rehabilitation training process based on a brain-computer interface into a time sequence arrangement structure, and establishing an inter-task-segment activation topology vector track.
2. The task state brain-computer interface training system for closed-loop transcranial magnetic stimulation according to claim 1, wherein the task segment partitioning module collects brain electrical signals and synchronous brain function activation records during the execution of a rehabilitation training task, and the construction of the task segment sequence specifically comprises: predefining a rehabilitation training flow based on a brain-computer interface and corresponding task event triggering conditions, and establishing a task event structure table; continuously acquiring electroencephalogram signals of different channels, adding a uniform time index, and constructing an electroencephalogram signal set with time alignment of each channel; Labeling a corresponding brain function activation tag when a task event is triggered, and presetting a stimulation strategy mapping rule for the brain function activation tag; And defining a paragraph boundary based on the time interval of the adjacent task event, integrating the electroencephalogram signals in the paragraph boundary with the brain function activation tags, and dividing the electroencephalogram signals into equal-length task segment sequences with set lengths.
3. The task state brain-computer interface training system for closed-loop transcranial magnetic stimulation according to claim 1, wherein the energy feature extraction module performs frequency domain conversion on brain electrical signals of different channels, and the extraction of the energy features of the brain electrical signals specifically comprises: screening electroencephalogram signal sets corresponding to different channels based on channel attribution from a task segment sequence; Executing a wavelet packet decomposition process on the electroencephalogram signals in each channel, and outputting a wavelet packet coefficient set according to a preset scale level; according to the frequency band distribution boundary related to cognition in the electroencephalogram research, merging components in the wavelet packet coefficient set into a component set of a set frequency band, and calculating the energy value in each frequency band; Reconstructing a local energy value sequence corresponding to each frequency band according to a signal sampling time axis by adopting a sliding window, establishing a time sequence energy curve, and generating a time sequence energy characteristic expression vector set covering a cognition related frequency band distribution boundary; and (3) arranging the time sequence energy characteristic expression vector set according to frequency bands to construct a two-dimensional energy matrix as a unified output structure of the electroencephalogram signal energy characteristics, wherein the dimensions of the two-dimensional energy matrix respectively correspond to the frequency band index and the sampling time point.
4. A task state brain-machine interface training system for closed-loop transcranial magnetic stimulation according to claim 1, wherein the identifier creation module uses the task activation topology vector and the brain function activation tag as training data, and creates an identification classifier of the brain function activation state comprises: extracting an activated topology vector track set from task activated topology vectors in a history training task, and acquiring a brain function activation tag carried by a task event to which a task segment corresponding to the topology vector track belongs; Acquiring a state transition candidate segment corresponding to a topological vector track, calculating a time difference between a corresponding starting time and a task event trigger time, and taking the time difference as brain function activation delay time; integrating the topological vector track sequence, the brain function activation delay time and the corresponding brain function activation tag, and constructing a structured training sample set by taking the task segment number corresponding to the topological vector track as a main index; and deploying a recognition classifier on the training sample set, and establishing a task brain function activation state recognition model.
5. The task state brain-machine interface training system for closed-loop transcranial magnetic stimulation according to claim 4, wherein the deploying the recognition classifier on the training sample set, and the establishing the task brain function activation state recognition model specifically comprises: setting a model structure of a brain function activation state identifier, and initializing a neural network weight matrix and an optimizer configuration parameter; performing a supervised learning process in the training sample set, taking the topological vector track and the brain function activation delay time as input features, and outputting a prediction result of a brain function activation tag; and carrying out gradient iteration update by adopting a cross entropy loss function and an Adam optimization algorithm, continuing iteration until the prediction error converges to the set index requirement, and storing the current model structure as a brain function activation state identifier.
6. The task state brain-machine interface training system for closed-loop transcranial magnetic stimulation according to claim 1, wherein the stimulation parameter recommendation module, by identifying the current brain function activation state, outputs transcranial magnetic stimulation recommendation parameters in real time specifically comprises: receiving a real-time electroencephalogram signal sequence of each channel, establishing a sliding time window by taking the length of a task segment as a sliding time window, and constructing a real-time topological vector track for the electroencephalogram signal sequence; Acquiring a time window starting point corresponding to the real-time topological vector track, calculating a trigger time difference between the time window starting point and a task event, inputting the trigger time difference into a brain function activation state identifier, and identifying the brain function activation state; based on the brain function activation state recognition result, inquiring the stimulation strategy mapping rule, and acquiring a corresponding stimulation rhythm, an action area and a stimulation parameter template as magnetic stimulation parameter recommendation input.

Description

Task state brain-computer interface training system for closed-loop transcranial magnetic stimulation Technical Field The invention relates to the technical field of cognitive rehabilitation assistance, in particular to a task state brain-computer interface training system for closed-loop transcranial magnetic stimulation. Background Transcranial magnetic stimulation is a conventional auxiliary means for cognitive repair, and in a task-driven transcranial magnetic stimulation brain-computer interface training system, external task flow structures and event trigger nodes are generally relied on as basis for stimulation time sequence control. Such systems generally assume that the test is in a corresponding cognitive functional state during a particular task phase and accordingly trigger transcranial magnetic stimulation during a set period of time. However, in practical application, the brain state to be tested often has obvious individual fluctuation and dynamic change in tasks, and a mode of accurately judging whether the brain state is in a stress intervention window or not is difficult to accurately judge by a trigger strategy which is simply dependent on a task section or event fixation, so that the problem of false triggering or missed triggering of stimulation can be caused, and the expected effect of transcranial magnetic stimulation on the regulation and control of the nerve functions of a trainer is difficult to obtain. In order to solve the above problems, a technical solution is now provided. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, embodiments of the present invention provide a task state brain-computer interface training system for closed-loop transcranial magnetic stimulation to solve the problems presented in the background art above. In order to achieve the above purpose, the present invention provides the following technical solutions: A task state brain-computer interface training system for closed-loop transcranial magnetic stimulation comprises a task segment dividing module, an energy feature extraction module, a state transition screening module, an activation topology construction module, a recognizer establishment module and a stimulation parameter recommendation module; the task segment dividing module collects brain electrical signals and synchronous brain function activation records during the execution of a rehabilitation training task, and constructs a task segment sequence; the energy characteristic extraction module performs frequency domain conversion on the electroencephalogram signals of different channels and extracts the energy characteristics of the electroencephalogram signals; the state transition screening module compares the frequency band energy distribution of the continuous task segments and screens state transition candidate segments; The activation topology construction module constructs a signal synchronization matrix based on time sequence brain signals of the state transition candidate segment, and analyzes task activation topology vectors crossing brain areas by using a graph convolution neural network; the identifier building module is used for building an identification classifier of the brain function activation state by taking the task activation topology vector and the brain function activation tag as training data; The stimulation parameter recommendation module outputs transcranial magnetic stimulation recommendation parameters in real time by identifying the current brain function activation state. In a preferred embodiment, the task segment dividing module collects electroencephalogram signals and synchronous brain function activation records during the execution of a rehabilitation training task, and the construction of the task segment sequence specifically includes: predefining a rehabilitation training flow based on a brain-computer interface and corresponding task event triggering conditions, and establishing a task event structure table; continuously acquiring electroencephalogram signals of different channels, adding a uniform time index, and constructing an electroencephalogram signal set with time alignment of each channel; Labeling a corresponding brain function activation tag when a task event is triggered, and presetting a stimulation strategy mapping rule for the brain function activation tag; And defining a paragraph boundary based on the time interval of the adjacent task event, integrating the electroencephalogram signals in the paragraph boundary with the brain function activation tags, and dividing the electroencephalogram signals into equal-length task segment sequences with set lengths. In a preferred embodiment, the energy feature extraction module performs frequency domain conversion on the electroencephalogram signals of different channels, and the extracting the energy features of the electroencephalogram signals specifically includes: screening electroencephalogram signal s