Rehabilitation training system based on TMS and VR cooperation of brain-computer interface

Abstract

The invention provides a brain-computer interface-based TMS and VR collaborative rehabilitation training system, which is characterized in that a multi-mode collaborative positioning module is used for collecting VR interaction data and electroencephalogram data and determining the optimal stimulation state of a subject; the motion intention analysis module is used for matching a rehabilitation mode according to the collected data and in combination with the type of the subject; and the closed-loop rehabilitation interaction module is used for adjusting parameters of the TMS equipment according to the rehabilitation mode and stimulating the final stimulation area under the optimal stimulation state of the subject. For a mild stroke patient, the system dynamically monitors the plastic change of the autonomic nerve and adaptively adjusts the treatment parameters according to the plastic change of the autonomic nerve; for a severe stroke patient, the virtual reality situation is combined to effectively induce motor imagery and synchronously collect and analyze electroencephalogram signals, closed-loop nerve regulation is achieved, synchronous accurate intervention and curative effect optimization of the severe stroke patient are achieved, and an intelligent solution is provided for rehabilitation treatment of the nerve function of the severe stroke patient.

Description

Technical Field

[0001] The present invention belongs to the fields of neural regulation technology and intelligent rehabilitation medical technology, and specifically relates to a rehabilitation training system based on TMS and VR collaboration based on a brain-computer interface. The system is suitable for personalized treatment of motor dysfunction in stroke patients, and is particularly suitable for critically ill patients who cannot be treated using clinical treatment plans due to the lack of electromyographic signals. Background Art

[0002] Transcranial magnetic stimulation (TMS) non-invasively regulates cerebral cortical excitability through magnetic fields and has been shown to promote neuroplasticity after stroke. Existing techniques often combine TMS with exercise training to enhance its therapeutic effects, but these techniques still have the following drawbacks:

[0003] Low spatiotemporal matching accuracy: The timing synchronization error between traditional TMS stimulation and patient movements exceeds 50ms, making it impossible to accurately activate the target neural circuit;

[0004] Severely ill patients have poor adaptability: approximately 20% of patients suffer from muscle atrophy or paralysis, resulting in a loss of electromyographic signals and inability to trigger stimulation through traditional closed-loop feedback mechanisms.

[0005] Static parameter setting: The existing system relies on fixed TMS parameters (such as frequency 10Hz, intensity 1.2T), which cannot be dynamically adjusted according to the patient's real-time neural activity, resulting in fluctuations in therapeutic efficacy. Summary of the Invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a rehabilitation training system based on TMS and VR collaboration based on brain-computer interface. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] A TMS and VR collaborative rehabilitation training system based on brain-computer interface includes: a multimodal collaborative positioning module, a movement intention analysis module and a closed-loop rehabilitation interaction module;

[0008] Multimodal collaborative positioning module, used to collect VR interaction data and EEG data, and determine the optimal stimulation state of the subject based on the EEG data;

[0009] The movement intention analysis module is used to match the rehabilitation mode based on the collected VR interaction data and original EEG data, combined with the subject's type;

[0010] The closed-loop rehabilitation interaction module is used for two-way communication with the VR device and the TMS device, adjusting the parameters of the TMS device according to the rehabilitation mode, and starting the TMS device to stimulate the final stimulation area under the optimal stimulation state of the subject.

[0011] Beneficial effects:

[0012] The present invention provides a rehabilitation training system based on TMS and VR collaboration of a brain-computer interface, wherein a multimodal collaborative positioning module is used to collect VR interaction data and EEG data and determine the optimal stimulation state of the subject; a movement intention analysis module is used to match the rehabilitation mode according to the collected data and in combination with the type of the subject; and a closed-loop rehabilitation interaction module is used to adjust the parameters of the TMS device according to the rehabilitation mode and stimulate the final stimulation area under the optimal stimulation state of the subject. For patients with mild stroke, the present invention dynamically monitors the changes in their autonomic neural plasticity and adaptively adjusts the treatment parameters accordingly; for patients with severe stroke, the present invention effectively induces motor imagery in combination with virtual reality scenarios, synchronously collects and analyzes EEG signals, and realizes closed-loop neural regulation, thereby achieving synchronous precise intervention and efficacy optimization for patients with mild or severe stroke, and providing an intelligent solution for the rehabilitation treatment of their neurological functions.

[0013] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Schematic diagram of the TMS and VR collaborative rehabilitation training system based on brain-computer interface provided by the present invention;

[0015] Figure 2 is a schematic diagram of the signal preprocessing provided by the present invention;

[0016] Figure 3 is a schematic diagram of image registration provided by the present invention;

[0017] Figure 4 It is a schematic diagram of the virtual coordinate system and 10-20 point display provided by the present invention.

[0018] Figure 5 is a schematic diagram of the UR3e robotic arm provided by the present invention;

[0019] Figures 6a-6e This is a simulation diagram of a 6-axis robotic arm provided by the present invention;

[0020] Figure 7 This is a schematic diagram of the connection of the acquisition MEP wires provided by the present invention;

[0021] Figure 8 This is a schematic diagram of EEG signal processing provided by the present invention;

[0022] Figure 9 This is the intent recognition flow chart provided by the present invention;

[0023] Figure 10 This is the architecture diagram of the motion intention analysis system based on VR and TMS provided by the present invention.

[0024] Figure 11 This is a schematic diagram of the TMS parameter control and VR feedback closed loop provided by the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0026] like Figure 1 As shown, the present invention provides a rehabilitation training system based on TMS and VR collaboration of brain-computer interface, including: a multimodal collaborative positioning module, a movement intention analysis module and a closed-loop rehabilitation interaction module;

[0027] Multimodal collaborative positioning module, used to collect VR interaction data and EEG data, and determine the optimal stimulation state of the subject based on the EEG data;

[0028] The movement intention analysis module is used to match the rehabilitation mode based on the collected VR interaction data and original EEG data, combined with the subject's type;

[0029] The closed-loop rehabilitation interaction module is used for two-way communication with the VR device and the TMS device, adjusting the parameters of the TMS device according to the rehabilitation mode, and starting the TMS device to stimulate the final stimulation area under the optimal stimulation state of the subject.

[0030] In a specific embodiment of the present invention, the multimodal collaborative positioning module is specifically used to:

[0031] S11, using VR equipment and EEG acquisition electrodes to collect VR interaction data and EEG data of the subjects;

[0032] S12, preprocessing the EEG data to obtain preprocessed EEG data;

[0033] In this step, a band-pass filter is used to band-pass filter the EEG data to obtain EEG data of a predetermined frequency band; independent component analysis is performed on the EEG data of the predetermined frequency band, and the electrooculogram and electromyography artifact data are removed, and then the data is segmented using a 1-2 second sliding window to obtain preprocessed EEG data.

[0034] refer to Figure 2As shown, MRI scans are performed on subjects to obtain structural or functional brain imaging data, while EEG is used to record EEG activity characteristics, such as alpha and beta waves. Joint analysis of EEG and MRI data can help verify the stability of brain states captured in MRI data. According to relevant research, alpha waves exhibit higher amplitudes and more stable frequencies when the brain is relaxed and awake. If EEG detects an abnormal decrease in alpha wave activity, it may indicate an overexcited brain state; conversely, significant fluctuations in alpha wave activity may indicate fatigue. In this case, the corresponding MRI data may be abnormal and require elimination.

[0035] During MRI scans, the subject's head inevitably moves slightly, which can cause motion artifacts in the image, seriously affecting the accuracy of subsequent analysis. This paper uses an advanced motion correction algorithm based on ITK (Insight Segmentation and Registration Toolkit) to precisely track the displacement of feature points in the image, accurately estimate head motion parameters, and perform targeted transformation correction on the image.

[0036] Image registration is also indispensable, and its core purpose is to achieve precise spatial alignment of images of different times, modalities, or individuals. The present invention can use a method that combines rigid transformation (Rigid Transformation) and affine transformation (AffineTransformation) for image registration. First, using rigid transformation, the images are initially aligned through translation and rotation operations; then, further optimization is performed with the help of affine transformation to ensure that the various structures in the image are accurately aligned in spatial position, providing a reliable data foundation for subsequent regional positioning.

[0037] S13, determining whether the amplitude of the pre-processed EEG data is the highest, and if so, determining that the subject is in the optimal stimulation state.

[0038] Based on the coordinate information of the target stimulation area, the stimulation area is accurately located with the help of structural or functional imaging data. For structural imaging data, the present invention uses standard brain maps such as ICBM152 [International Consortium for Brain Mapping (ICBM)] and MNI (Montreal Neurological Institute, Canada), and combines them with advanced image registration algorithms to determine the spatial position of the target area. Figure 3Using the ICBM152 standard spatial template as an example, the subject's MRI data is first registered with the ICBM152 template. During the registration process, affine and nonlinear transformations are applied sequentially to achieve optimal spatial alignment. Then, based on the coordinates of the target region in the standard atlas, the corresponding region is precisely located on the registered subject's image.

[0039] For functional imaging data, the location of the target region is determined by analyzing the statistical significance of activation. During functional MRI (fMRI) data processing, statistical analysis methods such as t-tests and F-tests are used to pinpoint brain regions that are significantly activated during specific tasks or states. For example, in fMRI studies involving cognitive tasks, by comparing brain activation before and after the task, regions whose activation exceeds a pre-set threshold are identified as the final stimulation regions.

[0040] refer to Figure 4 , marking the coordinate information of the initial stimulation area on the subject's scalp is an important prerequisite for achieving precise stimulation. First, based on the coordinates of the target area obtained earlier in the virtual coordinate system, combined with the principle of the individual 10-20 international brain electrode positioning system, the coordinate information is converted into the actual position on the scalp surface. Specifically, by accurately measuring the coordinate values ​​of the four positions of the nasal root (Nz), left ear root (AL), right ear root (AR) and internal occipital protuberance (Iz) of the head model reconstructed from MRI data in the virtual system, based on these reference points, complex calculations are performed to determine key positions such as the Cz point, and then the positions of each point in the 10-20 system on the scalp are accurately determined.

[0041] Then, using a professional marker pen or high-precision positioning stickers, the target stimulation area is precisely marked on the subject's scalp. These markings serve as a key reference for subsequent stimulation navigation and provide an important basis for the precise positioning of the transcranial magnetic stimulation coil.

[0042] refer to Figure 5 Based on the final stimulation location marker and the spatial position of the initial stimulation area, a high-precision, computer-controlled six-axis robotic arm holds the transcranial magnetic stimulation coil for precise positioning and stimulation. The computer control platform in the transcranial magnetic stimulation navigation system integrates data and instructions from various modules. It receives real-time MRI and EEG data from the image acquisition system, as well as head target position information acquired by the optical positioning system. After in-depth analysis and processing, it sends precise control instructions to the six-axis robotic arm.

[0043] UR3e robotic arm

[0044] MRI data acquired by the image acquisition system provides critical spatial information about the target area for positioning the robotic arm, while EEG data is used to monitor brain states in real time, providing a strong basis for precise selection of stimulation timing. The optical positioning system uses a binocular camera to track infrared reflective beads on the head support in real time. Using target tracking, target coordinate transformation, and head registration techniques, the target on the reconstructed head model is closely linked to the target on the real head, allowing for real-time and accurate positioning and tracking of the target on the real head. The six-axis robotic arm, upon receiving commands from the computer control platform, moves according to the robotic arm system control principles. First, coil calibration is performed to accurately determine the transformation relationship between the coil's tool center point (TCP) and the center of the robotic arm's end. Then, using the robotic arm inverse solution algorithm, the parameters of each joint axis of the robotic arm are calculated based on the coordinates of the target, enabling precise control of the robotic arm's motion. During motion planning, an improved RRT algorithm is used to ensure that the robotic arm moves safely, efficiently, and accurately along the pre-set trajectory to the stimulation target.

[0045] After reaching the target location, the robotic arm triggers the transcranial magnetic stimulation coil to produce stimulation. At the same time, the stimulation effect and positioning accuracy are verified by collecting electrophysiological data (such as MEP amplitude, latency, etc.) or behavioral data (such as the subject's behavioral performance, cognitive ability changes, etc.) of the stimulation response. Figure 7 For example, by collecting the MEP signal of the muscle after stimulation and comparing the MEP amplitude under different stimulation positions and stimulation parameters, the stimulation effect can be scientifically evaluated; by observing the behavioral changes of the subjects before and after stimulation, the accuracy of the positioning can be objectively judged, so as to continuously optimize the stimulation plan.

[0046] In a specific embodiment of the present invention, the motion intention analysis module is specifically used to:

[0047] S21, extracting EEG features of preprocessed EEG data and collecting VR interaction data of the subjects;

[0048] The raw EEG signals were bandpass filtered between 0.5 and 40 Hz to remove power frequency interference and myoelectric artifacts. Independent component analysis (ICA) was used to eliminate eye movement artifacts while retaining signals from movement-related brain regions (such as electrodes C3 / C4). The power spectral density of movement-related frequency bands (beta and theta waves) was calculated in real time using short-time Fourier transform (STFT), with the feature vector updated every 200 ms.

[0049] S22, integrating EEG features with VR interaction data, and identifying the subject's intention through a dynamic threshold algorithm; the intention recognition results include the presence of autonomous movement intention, the presence of VR-induced intention, and the absence of movement intention;

[0050] Using a dynamic threshold algorithm, spontaneous movement intention in mild patients was triggered by a sudden increase in beta wave power, while VR-induced movement imagery intention in severe patients was determined by the theta / beta power ratio (threshold ≥ 1.5). The raw EEG signal was filtered with a 50 Hz power frequency notch, and time-frequency analysis was performed using continuous wavelet transform (CWT) to extract event-related desynchronization (ERD) features in leads C3 / C4.

[0051] When a patient with severe brain injury wears an EEG acquisition device and initiates a virtual reality training program, the system initiates a multidimensional neural remodeling process. When the patient's visual focus is locked on a target object in the virtual reality scene (e.g., a three-dimensional teacup model), the desynchronized μ rhythm (8-12 Hz) generated by the premotor cortex is captured by the EEG electrode array, triggering the following cascaded rehabilitation intervention process: To address the signal-to-noise ratio (SNR) characteristics of critically ill patients (less than 3 dB), the system activates a deep residual network (ResNet-18) for feature space mapping, improving the accuracy of motor imagery classification from 68.2±5.1% to 82.7±4.3%. Motor commands, decoded by a support vector machine (SVM, radial basis kernel γ = 0.8), are transmitted to the virtual reality rendering engine via the ROS middleware.

[0052] The Unity3D engine dynamically generates rehabilitation scenarios based on historical patient training data (Fugl-Meyer scores and electromyographic signal amplitudes). The system's built-in physics engine (NVIDIA PhysX 5.1) calculates virtual object dynamic parameters in real time: when the patient's estimated grip strength is detected to be less than 10N, the teacup's mass parameter is automatically reduced from 300g to 150g, and the collision volume expansion coefficient is adjusted to 1.3 times. The tactile feedback device (TactGlove-MK3) outputs corresponding mechanical feedback based on the virtual material properties, and the silicone grip's stiffness coefficient is set to 0.8N / mm±0.05.

[0053] S23, matching a corresponding rehabilitation mode according to the recognition result of the subject's intention.

[0054] Patients wearing VR equipment and EEG acquisition electrodes generate VR interaction data and raw EEG signals, which are then synchronously transmitted to an embedded processor. The embedded processor integrates signal preprocessing, feature extraction, intent recognition, and TMS parameter generation modules. After processing and analyzing the data, it generates TMS parameter control instructions to drive the TMS stimulation device and simultaneously feeds the stimulation effects back into the VR scene, forming a closed-loop rehabilitation training system consisting of "VR interaction - EEG acquisition - intent analysis - TMS stimulation - VR feedback."

[0055] In a specific embodiment of the present invention, extracting EEG features of pre-processed EEG data and collecting VR interaction data of the subject include:

[0056] S31, for each pre-processed EEG data in the sliding window, calculating the power spectrum density of the pre-processed EEG data and using it as an EEG feature;

[0057] When patients perform motor imagery or interactive tasks in a VR scene, EEG electrodes collect raw EEG signals from motor cortical areas such as C3 and C4 at a sampling rate of no less than 256Hz. The signals are first band-pass filtered to extract the beta wave (13-30Hz) and theta wave (4-8Hz) frequency band signals, and then subjected to independent component analysis (ICA) to remove electrooculogram and electromyographic artifacts before being segmented into 1-2 second sliding windows (with an overlap rate of 50%) to complete preprocessing. At the same time, interactive data such as hand movement speed, acceleration, and force feedback intensity of the force feedback device collected by the VR data gloves are extracted and merged with the PSD features into a multi-dimensional feature vector as input for motion intention recognition.

[0058] For the β / θ wave signal in each sliding window, the Welch algorithm is used to calculate the power spectral density (PSD). The calculation formula of the power spectral density is expressed as:

[0059]

[0060] Where x(n) is the signal in the window, ω(n) is the Hanning window function, N is the number of window length points, f is the frequency, W is the Hanning window normalization coefficient, and n is the discrete sampling point number.

[0061] S32, calculating the mean and variance of the power spectrum density, and setting a dynamic threshold interval according to the mean and variance;

[0062] Before the patient begins training, resting-state EEG data are collected, and the mean μ and standard deviation σ of the β / θ wave PSD are calculated to set the dynamic threshold interval, expressed as: [μ-2σ, μ+2σ], where μ represents the mean of the power spectral density and σ represents the variance of the power spectral density. During the training process, after every 10 sliding window calculations, the threshold is updated according to the new data using a formula. For the EEG features within each sliding window, the mean and variance of the power spectral density are updated in real time, thereby updating the dynamic threshold interval.

[0063] The update formula of the mean value of the power spectral density is expressed as:

[0064]

[0065] Where μ new represents the mean of the updated power spectral density, α represents the forgetting factor, μ old represents the mean of the power spectral density before updating, Represents the mean of the power spectral density before updating.

[0066] The update formula of the variance of the power spectral density is expressed as:

[0067]

[0068] Where, σ new represents the variance of the updated power spectral density, σ new represents the variance of the power spectral density before updating, x i Represents the power spectrum density data sample corresponding to the i-th EEG data.

[0069] S33, collecting the subject's hand movement speed, acceleration and virtual displacement, and using the hand movement speed, acceleration and virtual displacement as VR interaction data.

[0070] In a specific embodiment of the present invention, the EEG features are integrated with VR interaction data, and the intention recognition result of the subject is identified by a dynamic threshold algorithm, including:

[0071] S41, for any subject, if the power spectrum density corresponding to the subject is not within the dynamic threshold range and the subject is a mild patient, then determine whether the VR interaction data of the subject meets the movement conditions. If so, determine that the subject has autonomous movement intention;

[0072] S42: For any subject, if the power spectral density of the subject is not within the dynamic threshold range and the subject is a critically ill patient, then the VR interaction data of the subject is judged to not meet the motion condition, and the subject is judged to have VR-induced intention;

[0073] S43 , for any subject, if the power spectrum density of the subject is within the dynamic threshold range, it is determined that the subject has no intention to exercise.

[0074] The intention discrimination module integrates EEG signals (β / θ wave power spectral density, θ / β ratio) and VR interaction data (hand movement speed, acceleration, virtual displacement, etc.), and distinguishes three types of intentions through a dynamic threshold algorithm: for patients with mild symptoms, a sudden increase in β wave PSD and VR detection of real movements (such as speed / acceleration meeting the standards, virtual displacement ≥5cm) are judged as autonomous movement intentions, triggering high-frequency rTMS and increasing the difficulty of the task; for patients with severe symptoms, a θ / β ratio ≥1.5 and no real movements but VR virtual guidance are judged as VR-induced intentions, triggering low-frequency rTMS and adapting to the complexity of the scene; if the feature does not exceed the threshold or the interaction data does not meet the standards, it is judged as no intention / false triggering, no stimulation is performed, and the threshold is updated in real time. When the abnormality persists, medical staff are prompted to adjust the plan, realizing accurate closed-loop intention recognition and regulation for patients with mild and severe diseases.

[0075] In a specific embodiment of the present invention, matching a corresponding rehabilitation mode according to the subject's intention recognition result includes:

[0076] S51, if the subject's intention recognition result is that there is an intention to move voluntarily, the matching rehabilitation mode is: output 15-20 Hz high-frequency rTMS, and dynamically adjust the stimulation intensity within 80-110% MT according to the decrease in the power spectral density of the EEG data;

[0077] S52, if the subject's intention recognition result is that there is an intention to move voluntarily, the matching rehabilitation mode is: output low-frequency rTMS of 1-5 Hz, and dynamically adjust the stimulation intensity between 60-90% MT according to the decrease in the power spectral density of the EEG data;

[0078] S53, if the intention recognition result of the subject is that there is no movement intention, the matching rehabilitation mode is: no rTMS output.

[0079] According to the intent recognition results, Figure 8 The mapping rules shown here regulate TMS parameters. When a voluntary movement intention is identified, a high-frequency rTMS of 15-20 Hz is output, and the stimulation intensity is dynamically adjusted between 80-110% MT based on the decrease in PSD. If the intention is VR-induced, a low-frequency rTMS of 1-5 Hz is used, with the intensity set to 60-90% MT. The stimulation duration matches the duration of the VR task and does not exceed 30 seconds. At the same time, the system sets the intensity cap at 120% MT and monitors the patient's heart rate and electromyographic signals in real time, automatically pausing stimulation if any abnormality is detected.

[0080] In a specific embodiment of the present invention, the closed-loop rehabilitation interaction module is specifically used to:

[0081] Two-way communication with VR devices and TMS devices;

[0082] Start the TMS device and adjust the parameters of the TMS device so that it outputs the corresponding rTMS signal according to the stimulation intensity and frequency parameters in the rehabilitation mode, and stimulate the final stimulation area under the subject's optimal stimulation state, thereby completing the subject's training process;

[0083] After the training is completed, an optimization report is generated and fed back to the VR device.

[0084] refer to Figure 9 After TMS stimulation, changes in cortical excitability are fed back into the VR scene, adjusting the virtual character's movement smoothness and task reward mechanisms based on the stimulation effect. For example, when the stimulation is effective, the virtual character's movements become more agile; otherwise, they become sluggish. After each training session, the system combines the patient's rehabilitation assessment results (such as the Fugl-Meyer score) to optimize the VR scene difficulty curve and TMS parameter mapping model, achieving continuous optimization of the rehabilitation training system.

[0085] The transcranial magnetic stimulation (TMS) parameter dynamic control module implements closed-loop optimization based on a deep reinforcement learning model (PPO algorithm). When the virtual hand model collides with the target object, the system outputs a monophasic pulse within a time window of 122.3±0.8ms (stimulation intensity 90-110% of resting motor threshold, pulse interval 180-250ms). The stimulation target is corrected in real time using the Neuronavigation system to ensure that the positioning error between the coil center and the primary motor cortex (M1 area) is less than 1.2mm.

[0086] Multi-source data generated during training (EEG time-frequency characteristics, virtual trajectory offsets, and TMS-induced MEP amplitudes) is uploaded in real time to a cloud-based analysis platform via an Apache Kafka pipeline. A rehabilitation efficacy prediction model based on the XGBoost algorithm updates its parameters every 24 hours, dynamically adjusting the training difficulty gradient for the following day. The digital twin subsystem integrates DTI fiber tracking data with resting-state fMRI functional connectivity matrices, quantitatively demonstrating an average daily increase of 0.15±0.03 in the fractional anisotropy (FA) of the corticospinal tract and an 18.7±2.5% increase in the Granger causal connectivity strength of the motor network.

[0087] This system adopts a three-level interlocking protection design:

[0088] Physiological signal monitoring layer: Continuous excess of delta wave power (>40% of total power) triggers training pause.

[0089] Equipment safety layer: The air cooling system is activated when the TMS coil temperature is greater than 40°C, and the stimulation intensity is automatically reset if it exceeds the limit.

[0090] Motion compensation inhibition: A visual warning prompt is activated when the EMG activity of the healthy side is greater than 300% of the affected side.

[0091] This closed-loop system achieves neuroplasticity regulation for motor function reconstruction through EEG-TMS-VR multimodal information fusion. Clinical validation data show that after 8 weeks of intervention, the upper limb Fugl-Meyer score of critically ill patients increased by 29.6±4.2 points (p<0.001), and the motor cortex activation volume expanded by 2.3±0.7cm. 3 (fMRI, FDR-corrected p<0.05), confirming the significant effectiveness of this technology in promoting neural network reorganization. Overall system workflow.

[0092] 1. Patient classification and initialization: Patients are classified as mild or severe based on clinical assessment results, and RMT baseline values ​​and personalized EEG baseline parameters are recorded.

[0093] 2. Data Acquisition and Synchronization: EEG signals were continuously acquired via a multimodal co-localization module, and TMS stimulation was triggered within 0-2 ms after detecting movement intention. EEG signals were sampled at 500 Hz and stored for offline analysis.

[0094] 3. Closed-loop control and iteration: After each treatment session (30 minutes), the system generates an optimization report based on EEG feedback data (such as intention recognition accuracy and virtual task completion), and automatically adjusts the TMS parameters and VR task configuration for the next treatment session.

[0095] This application proposes a rehabilitation training system and method based on brain-computer interface-based transcranial magnetic stimulation and virtual reality collaboration, aiming to promote the widespread application of transcranial magnetic stimulation technology in the field of rehabilitation treatment for stroke patients. The system adopts a high-precision 6-axis robotic arm positioning device to achieve sub-millimeter spatial alignment of transcranial magnetic stimulation (TMS) and electroencephalogram (EEG) signals, and integrates deep reinforcement learning algorithms to build a dynamic optimization model for stimulation parameters. The system adopts a modular individualized design concept to provide differentiated rehabilitation plans for patients at different stages of the disease: for patients with mild stroke, the system dynamically monitors the changes in their neuroplasticity during autonomous rehabilitation training and adaptively adjusts the treatment parameters accordingly; for patients with severe stroke, it combines virtual reality scenario simulation technology to effectively induce motor imagery, synchronously collects and analyzes EEG signals, and realizes closed-loop neural regulation. By establishing a precise mapping relationship between individualized parameter control and rehabilitation training, this system innovatively achieves synchronous precise intervention and efficacy optimization for patients with mild and severe stroke, providing an intelligent solution for their neurological rehabilitation treatment.

[0096] It is worth noting that the terms "first" and "second" in this disclosure are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more of such features. In the description of this disclosure, "plurality" means two or more, unless otherwise specifically defined.

[0097] Although the present application is described herein with reference to various embodiments, those skilled in the art will be able to understand and implement other variations of the disclosed embodiments in practicing the claimed application by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

[0098] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A rehabilitation training system based on TMS and VR collaboration based on brain-computer interface, characterized in that: include: Multimodal collaborative localization module, motion intention parsing module, and closed-loop rehabilitation interaction module; The multimodal collaborative positioning module is used to collect VR interaction data and EEG data, and determine the optimal stimulation state of the subject based on the EEG data; The movement intention analysis module is used to match the rehabilitation mode according to the collected VR interaction data and original EEG data and the subject's type; The closed-loop rehabilitation interaction module is used to communicate bidirectionally with the VR device and the TMS device, adjust the parameters of the TMS device according to the rehabilitation mode, and start the TMS device to stimulate the final stimulation area under the optimal stimulation state of the subject.

2. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 1 is characterized in that: The multimodal collaborative positioning module is specifically used to: Use VR equipment and EEG acquisition electrodes to collect VR interaction data and EEG data of the subjects; Preprocessing the EEG data to obtain preprocessed EEG data; It is determined whether the amplitude of the pre-processed EEG data is the highest. If so, it is determined that the subject is in the optimal stimulation state.

3. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 2 is characterized in that: The preprocessing of the EEG data to obtain preprocessed EEG data comprises: Using a bandpass filter to perform bandpass filtering on the EEG data to obtain EEG data of a predetermined frequency band; Independent component analysis is performed on the EEG data of the predetermined frequency band, and the electrooculogram and electromyography artifact data are removed, and then the data is segmented using a sliding window of 1-2 seconds to obtain pre-processed EEG data.

4. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 2 is characterized in that: The motion intention analysis module is specifically used to: Extracting EEG features of the preprocessed EEG data and collecting VR interaction data of the subject; The EEG features are integrated with the VR interaction data, and the intention recognition result of the subject is identified through a dynamic threshold algorithm; wherein the intention recognition result includes the presence of autonomous movement intention, the presence of VR-induced intention, and the absence of movement intention; The corresponding rehabilitation mode is matched according to the recognition results of the subject's intention.

5. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 4 is characterized in that: Extracting EEG features of the pre-processed EEG data and collecting VR interaction data of the subject include: For the pre-processed EEG data in each sliding window, the power spectrum density of the pre-processed EEG data is calculated and used as the EEG feature; Calculating the mean and variance of the power spectral density, and setting a dynamic threshold interval according to the mean and variance; The subject's hand movement speed, acceleration and virtual displacement are collected and used as VR interaction data.

6. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 5 is characterized in that: The result of fusing the EEG features with the VR interaction data and identifying the subject's intention through a dynamic threshold algorithm includes: For any subject, if the power spectral density corresponding to the subject is not within the dynamic threshold range and the subject is a mild patient, then it is determined whether the subject's VR interaction data meets the movement conditions. If so, it is determined that the subject has autonomous movement intention; For any subject, if the power spectral density of the subject is not within the dynamic threshold range and the subject is a critically ill patient, then the VR interaction data of the subject is judged to not meet the motion conditions, and the subject is determined to have VR-induced intention; For any subject, if the power spectral density of the subject is within the dynamic threshold range, it is determined that the subject has no movement intention.

7. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 6 is characterized in that: The calculation formula of the power spectrum density is expressed as: Where x(n) is the signal in the window, ω(n) is the Hanning window function, N is the number of window length points, f is the frequency, W is the Hanning window normalization coefficient, and n is the discrete sampling point number; The dynamic threshold interval is expressed as: [μ-2σ, μ+2σ], where μ represents the mean of the power spectrum density, and σ represents the variance of the power spectrum density.

8. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 7 is characterized in that: For the EEG features within each sliding window, the mean and variance of the power spectrum density are updated in real time, thereby updating the dynamic threshold interval; The update formula of the mean value of the power spectrum density is expressed as: Where μ new represents the mean of the updated power spectral density, α represents the forgetting factor, μ old represents the mean of the power spectral density before updating, represents the mean of power spectral density; The update formula of the variance of the power spectral density is expressed as: Where, σ new represents the variance of the updated power spectral density, σ new represents the variance of the power spectral density before updating, x i Represents the power spectrum density data sample corresponding to the i-th EEG data.

9. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 4 is characterized in that: The corresponding rehabilitation modes matched according to the subject's intention recognition results include: If the subject's intention recognition result indicates that there is an intention to move autonomously, the matching rehabilitation mode is: output 15-20 Hz high-frequency rTMS, and dynamically adjust the stimulation intensity within 80-110% MT according to the decrease in the power spectral density of the EEG data; If the subject's intention recognition result indicates that there is an intention to move autonomously, the matching rehabilitation mode is: output low-frequency rTMS of 1-5 Hz, and dynamically adjust the stimulation intensity between 60-90% MT according to the decrease in the power spectral density of the EEG data; If the subject's intention recognition result is that there is no movement intention, the matching rehabilitation mode is: no rTMS output.

10. The TMS and VR collaborative rehabilitation training system based on brain-computer interface according to claim 1, characterized in that: The closed-loop rehabilitation interaction module is specifically used to: Two-way communication with VR devices and TMS devices; Start the TMS device and adjust the parameters of the TMS device so that it outputs the corresponding rTMS signal according to the stimulation intensity and frequency parameters in the rehabilitation mode, and stimulate the final stimulation area under the subject's optimal stimulation state, thereby completing the subject's training process; After the training is completed, an optimization report is generated and fed back to the VR device.

Images