Post-stroke upper limb sensory motor function monitoring method and system based on multi-task state OPM-MEG differential decoupling

Abstract

This invention belongs to the field of medical and health care rehabilitation monitoring technology, and provides a method and system for monitoring upper limb sensory and motor function after stroke based on multi-task OPM-MEG differential decoupling. The method includes acquiring brain magnetic signal data, stimulus data, and behavioral state data of the target subject under different tasks at different monitoring time points; mapping these data to obtain a spatiotemporal sequence of brain-derived activity and extracting a joint feature vector; obtaining a passive control feature vector through active-passive trial matching; performing multi-task feature differential decoupling processing to obtain sensory afferent representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters; and generating longitudinal monitoring results through longitudinal comparison. The method and system for monitoring upper limb sensory and motor function after stroke based on multi-task OPM-MEG differential decoupling provided by this invention reduces the error of cross-treatment time analysis results through differential decoupling of multi-task data, thereby improving the accuracy and reliability of longitudinal monitoring results.

Description

Technical Field

[0001] This invention belongs to the fields of brain magnetic signal processing, neurofunctional imaging and medical and health care rehabilitation monitoring technology, and specifically relates to a method and system for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling. Background Technology

[0002] In clinical practice, stroke often leads to varying degrees of upper limb motor dysfunction. Patients typically exhibit limited finger flexion and extension, decreased grip strength, unstable wrist control, and incoordination between voluntary movement and sensory feedback. During rehabilitation, scales or grip strength measurements, range of motion measurements, and task completion time are commonly used to evaluate upper limb motor function. While these indicators can reflect the patient's external motor performance, they are often influenced by factors such as the assessor's experience, the patient's cooperation, and compensatory movements. Furthermore, these indicators cannot directly reflect changes in the patient's central motor commands, peripheral sensory input, and sensorimotor integration function.

[0003] Currently, while electroencephalography (EEG) can record neural activity with high temporal resolution, its spatial localization is easily affected by the conductivity of tissues such as the scalp and skull. Functional magnetic resonance imaging (fMRI), while providing good spatial resolution, has relatively low temporal resolution, and patients typically need to remain as still as possible during the scan, which is clearly unfavorable for performing natural upper limb movement tasks. Furthermore, traditional magnetoencephalography (MEG) systems based on superconducting quantum interference devices (SQUIDs) require cryogenic operation, and the corresponding sensor arrays are fixed within a Dewar container, resulting in a large distance between the sensors and the scalp and making it difficult to accommodate patients with different head shapes. In contrast, the optically pumped magnetometer (OPM-MEG) system uses an OPM-MEG system, which employs an OPM-MEG meter that can be placed close to the scalp to measure the weak magnetic field generated by neural currents. It achieves both high temporal resolution and source localization capabilities, and allows for controlled motor imagery, motor attempts, and actual motor tasks. Therefore, OPM-MEG provides new technical conditions for the objective monitoring of upper limb sensory and motor function after stroke.

[0004] However, existing brain function rehabilitation monitoring programs generally use resting-state data, single active motor tasks, or single motor imagery tasks to extract indicators such as brain region activation intensity, event-related desynchronization, and brain region connectivity. On the one hand, responses collected in active motor states simultaneously include multiple components such as motor planning, motor commands, muscle output, tactile and proprioceptive feedback, and sensorimotor integration; on the other hand, passive motor states mainly include sensory input caused by external movement, but they are still affected by factors such as attention, range of motion, and passive load differences; furthermore, while motor imagery states can reflect a certain degree of motor planning, some patients may exert unexpected force during the imagery process. Obviously, if only a single task or different unmatched tasks are compared, the rehabilitation monitoring results are difficult to accurately distinguish whether the improvement in the patient's external movement comes from the recovery of central motor commands, improvement in sensory feedback, or increase in compensatory activity, making the rehabilitation monitoring results unable to truly reflect the patient's rehabilitation status.

[0005] Therefore, there is an urgent need to design a method and system for monitoring upper limb sensorimotor function after stroke that can integrate multi-task monitoring data of resting, motor imagery, active movement and passive movement, and perform task matching and longitudinal comparison. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a method and system for monitoring upper limb sensory and motor function after stroke based on multi-task OPM-MEG differential decoupling. By differentially decoupling multi-task data, the influence of factors such as different sensor positions, different levels of task completion, and differences in patient passive feedback in cross-treatment time comparison analysis is reduced, thereby improving the accuracy and reliability of longitudinal monitoring results. This solves the problems in existing rehabilitation monitoring methods, such as the difficulty in separating sensory feedback from central motor commands from active motor magnetoencephalography responses and the inconsistency of biomechanical conditions in different tasks.

[0007] The technical solution of this invention is:

[0008] This invention proposes a method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling, comprising the following steps:

[0009] Step S100: Collect and obtain the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target object when performing resting task, upper limb motor imagery task, upper limb active motor task, and upper limb passive motor task at at least two monitoring time points respectively;

[0010] Step S200: Perform data preprocessing and synchronization alignment processing on the brain magnetic signal data, stimulus data, and behavioral state data at each monitoring time point, and combine the stimulus data and behavioral state data to determine the actual movement time and thereby divide the task trials into segments.

[0011] Step S300: Map the brain magnetic signal data collected at each monitoring time point under each task to a unified brain source space based on an individualized head model or a standard head model, and generate a brain source activity spatiotemporal sequence corresponding to each task.

[0012] Step S400: Construct behavioral feature vectors for upper limb active movement task and upper limb passive movement task trials based on behavioral state data and task trial segmentation, and perform active-passive trial matching processing on the two to obtain a passive control feature vector corresponding to the behavioral feature vector of the upper limb active movement task trial.

[0013] Step S500: Based on the spatiotemporal sequence of brain source activity for each task, construct a multi-brain region network covering the sensorimotor brain regions on the same side and opposite side of the lesion, and extract the joint feature vector for each task. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features.

[0014] Step S600: Based on the joint feature vectors corresponding to the resting task, upper limb motor imagery task, upper limb active motor task, and upper limb passive motor task, and the passive control feature vector corresponding to the upper limb active motor task, perform multi-task state feature difference decoupling processing to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters.

[0015] Step S700: Generate longitudinal monitoring results based on the sensory input representation parameters, the central motor command representation parameters, and the sensorimotor integration representation parameters at different monitoring time points.

[0016] Preferably, the active-passive trial matching process includes the following steps:

[0017] Step S410: Calculate the behavioral distance between the behavioral feature vectors of the upper limb active movement task trials and the behavioral feature vectors of the upper limb passive movement task trials.

[0018] Step S420: Combine the behavioral distance and the preset matching threshold to filter and generate a candidate matching set that corresponds to the upper limb active movement task trials and consists of upper limb passive movement task trials;

[0019] Step S430: If the candidate matching set contains only one upper limb passive movement task trial, then establish a one-to-one pairing and use the behavioral feature vector of the upper limb passive movement task trial as the passive control feature vector corresponding to the current upper limb active movement task trial.

[0020] If the candidate matching set contains multiple upper limb passive movement task trials, then normalized weights are generated based on the behavioral distance between each upper limb passive movement task trial and the upper limb active movement task trial, and the behavioral feature vectors of multiple upper limb passive movement task trials are weighted and averaged to obtain the passive control feature vector corresponding to the upper limb active movement task trial.

[0021] Preferably, the multi-task-state feature difference decoupling process includes the following steps:

[0022] Step S610: Generate the sensory input representation parameters by differential calculation based on the joint feature vector of the resting task and the joint feature vector of the upper limb passive movement task;

[0023] Step S620: Calculate and generate the central motor instruction representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector.

[0024] Step S630: Calculate and generate the sensorimotor integration representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector.

[0025] Preferably, the calculation and acquisition of the central motor command representation parameters includes the following steps:

[0026] Step S621: Calculate the differential features of the movement plan based on the joint feature vector of the upper limb motor imagery task and the joint feature vector of the resting task;

[0027] Step S622: Calculate the active participation differential features based on the joint feature vector of the upper limb active movement task and the passive control feature vector;

[0028] Step S623: The central motor command representation parameters are obtained by weighted calculation based on the differential features of the movement plan and the differential features of active participation.

[0029] Preferably, the calculation and acquisition of the sensorimotor integration representation parameters includes the following steps:

[0030] Step S631: Calculate the predicted active feature vector based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, and the passive control feature vector of the upper limb active motor task.

[0031] Step S632: Calculate the sensorimotor integration representation parameters based on the predicted active feature vector and the joint feature vector of the upper limb active movement task.

[0032] Preferably, the longitudinal monitoring results include the trajectory of changes in sensory afferent representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters, as well as the affected-unaffected side comparison results and health reference standard scores.

[0033] Preferably, the present invention also provides a post-stroke upper limb sensory and motor function monitoring system based on multi-task state OPM-MEG differential decoupling, for performing the aforementioned method, including:

[0034] The data acquisition module is used to collect the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target object when performing resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks at at least two monitoring time points.

[0035] The data processing module is used to preprocess and synchronize the brain magnetic signal data, stimulus data, and behavioral state data at each monitoring time point, and to determine the actual movement time by combining the stimulus data and behavioral state data, and thereby divide the task trials into segments.

[0036] The source space mapping module is used to map the magnetoencephalogram (MEG) signal data collected at each monitoring time point under each task to a unified brain source space based on an individualized head model or a standard head model, and generate a spatiotemporal sequence of brain source activity corresponding to each task.

[0037] The matching processing module is used to construct behavioral feature vectors for upper limb active movement tasks and upper limb passive movement tasks based on behavioral state data and task trial segmentation, and to perform active and passive trial matching processing on the two to obtain a passive control feature vector corresponding to the behavioral feature vector of the upper limb active movement task trial.

[0038] The feature extraction module is used to construct a multi-brain region network covering the ipsilateral and contralateral sensorimotor brain regions of the lesion based on the spatiotemporal sequence of brain source activity for each task, and to extract the joint feature vector for each task. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features.

[0039] The differential decoupling module is used to perform multi-task state feature differential decoupling processing to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters based on the joint feature vectors corresponding to resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks, as well as the passive control feature vector corresponding to upper limb active motor tasks.

[0040] The longitudinal monitoring module is used to generate longitudinal monitoring results based on sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters at different monitoring time points.

[0041] The present invention has the following advantages and effects compared with the prior art:

[0042] (1) A multi-task joint comparative analysis was adopted, including resting task, upper limb motor imagery task, upper limb active motor task and upper limb passive motor task. Compared with the existing monitoring method that uses active motor magnetic response as a single motor function indicator, this method comprehensively monitors the upper limb sensorimotor function indicators by integrating multiple representation parameters such as sensory input, central motor command and sensorimotor integration, thereby improving the accuracy and reliability of longitudinal monitoring results and making the monitoring results have a clearer neurophysiological meaning.

[0043] (2) Active-passive trial matching processing is used to pair or weight the behavioral feature vectors corresponding to the active and passive upper limb movement tasks to obtain passive control feature vectors. This is used to reduce the impact of differences between active and passive tasks in terms of parameters such as movement load, movement amplitude, duration, and movement curve on subsequent EEG differential results, thereby reducing the bias and error introduced therefrom.

[0044] (3) Using a unified brain source space based on an individualized head model or a standard head model, the brain magnetic signal data at different monitoring time points are mapped to a consistent unified brain source space, reducing the impact of positional differences caused by repeated wearing of OPM-MEG system sensors on the monitoring results, and meeting the actual clinical need for longitudinal monitoring and rehabilitation analysis across treatment courses.

[0045] (4) The sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters are obtained by multi-task characteristic difference decoupling processing. The longitudinal monitoring results are comprehensively and specifically displayed by the longitudinal changes of multiple functional representation parameter components at different monitoring time points. This helps to distinguish the changes of different neural functional components, reduce the result bias caused by low-quality data, and improve the reliability of monitoring results. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the post-stroke upper limb sensory and motor function monitoring method based on multi-task state OPM-MEG differential decoupling in Embodiment 1 of the present invention.

[0047] Figure 2 This is a schematic diagram illustrating the principle of the post-stroke upper limb sensory and motor function monitoring method based on multi-task state OPM-MEG differential decoupling in Embodiment 1 of the present invention.

[0048] Figure 3 This is a schematic diagram of the architecture of the post-stroke upper limb sensory and motor function monitoring system based on multi-task state OPM-MEG differential decoupling in Embodiment 2 of the present invention. Detailed Implementation

[0049] To enable those skilled in the art to better understand the present invention, specific embodiments will now be described in further detail. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention.

[0050] Example 1:

[0051] like Figure 1 and Figure 2 As shown, this invention provides a method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling, and particularly relates to a method for monitoring upper limb sensory and motor function after stroke targeting subjects with upper limb motor dysfunction after stroke, which specifically includes the following steps:

[0052] Step S100: Collect and acquire multi-task-state data from multiple monitoring time points:

[0053] Collect and acquire the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target subject when performing resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks at at least two monitoring time points.

[0054] Specifically, behavioral state data includes data that can be used to describe the kinematic states of the upper limbs, such as force state, displacement state, posture state, velocity state, and acceleration state, and is acquired in real time through corresponding sensors.

[0055] It should be noted that the monitoring time points mentioned here specifically refer to an independent OPM-MEG examination during the rehabilitation process, which includes, but is not limited to, the baseline time point before the start of rehabilitation, one or more follow-up time points during the rehabilitation process, and the assessment time point after the end of the rehabilitation phase.

[0056] Furthermore, the upper limb motor imagery task here specifically refers to the imagery motor task in which the target object completes a preset upper limb movement solely through imagination without actually moving; the upper limb active motor task specifically refers to the autonomous motor task in which the target object can produce visible movement or measurable mechanical output, as well as the motor attempt task in which the target object attempts to move according to instructions but only produces a weak mechanical output; the upper limb passive motor task specifically refers to the non-autonomous motor task in which the target object does not actively exert force and the upper limb is driven by an actuator to complete the movement.

[0057] Specifically, in this embodiment, the target object performs resting tasks, upper limb motor imagery tasks, upper limb active movement tasks, and upper limb passive movement tasks at the baseline monitoring time point and subsequent monitoring time points, respectively. At the same time, multi-channel magnetoencephalography (MEG) data during the execution of the above four types of tasks are collected and recorded by the OPM-MEG (Optically Pumped Magnetometer Magnetoencephalography) system. Stimulus data is collected and recorded synchronously by the stimulus presentation system (such as a virtual reality headset or projection device). Upper limb behavioral state data is collected and recorded synchronously by the behavioral state system (such as a data acquisition system integrating torque sensors, joint angle sensors, inertial measurement units, or optical motion capture units). The OPM-MEG system, stimulus presentation system, and behavioral state system use a shared clock, or synchronization pulse, or calibrable timestamp to establish a unified time reference for data synchronization.

[0058] The following is a detailed explanation using a target object grasping experiment as an example:

[0059] The target subject sits on a non-magnetic chair in a magnetically shielded environment, or lies on a non-magnetic examination bed in a magnetically shielded environment. The target subject's forearm is supported by a non-magnetic support structure, and the affected hand naturally holds a grip strength meter. The grip strength meter outputs grip strength timing data through a data acquisition interface that is wired isolated, fiber optic isolated, or located outside the shielded environment.

[0060] Each trial consists of a baseline phase, a cueing phase, a task phase, and a recovery phase. The baseline phase estimates the target subject's pre-task magnetic resonance imaging (MRI) power and baseline muscle tone. The cueing phase provides cues for the movement category and the affected hand. The task phase involves performing resting tasks, upper limb motor imagery tasks, active upper limb movement tasks (active grasping dynamometer), and passive upper limb movement tasks (passive grasping dynamometer). The recovery phase retrieves post-movement information. Rebound and brain region recovery timeline.

[0061] It should be noted that the duration of the baseline phase, cue phase, task phase, and recovery phase are uniformly set according to the tolerance level of the target subjects, and the same task structure and stimulus labeling rules are used for the same target subjects at different monitoring time points.

[0062] In the upper limb motor imagery task during the task phase, the target subject keeps the affected hand relaxed and imagines performing the same action as the active grasping action in the upper limb active motor task. The grip strength meter continuously records the mechanical output, and when the grip strength continuously exceeds the threshold determined based on the resting period distribution, grip strength meter resolution, and baseline muscle tone, the corresponding trial is removed or remarked as an upper limb active motor task trial. This avoids mistaking the EEG response generated by unexpected actual force exertion as a pure imagined motor response.

[0063] In the upper limb active movement task during the task phase, the target subject's maximum safe grip strength is measured first, and then a fixed-load task and a relative-load task are set accordingly. The fixed-load task uses the target grip strength determined at the baseline monitoring time point to compare changes in brain activity under the same peripheral output conditions; the relative-load task determines the target grip strength according to a preset proportion of the maximum safe grip strength at the current time point to compare changes in brain activity under similar individual load levels. Throughout this process, the actual grip strength curve is preserved, and the stimulus cue time is not used to replace the actual force exertion time.

[0064] In the passive upper limb movement task during the task phase, a passive actuator controls the affected hand of the target object to perform the same action as the active grasping action in the active upper limb movement task. The passive actuator moves the affected hand to complete the grasping and releasing action in the same direction and duration as the active grasping action, and outputs data such as actual force, displacement, pressure, actuator status, and control timing. It should be noted that the active-passive trial matching is performed based on the actual recorded values, and the set values ​​of the actuator do not directly replace the actual movement state.

[0065] In addition, passive actuators include, but are not limited to, any one or more of the following: artificial assistive mechanisms, non-magnetic mechanical mechanisms, actuation mechanisms, elastic traction mechanisms, rehabilitation gloves, upper limb rehabilitation robots, and exoskeleton mechanisms.

[0066] Taking this target object grasping experiment as an example, the behavioral state data specifically includes at least one or more of the following data collected by a grip strength meter: peak grip strength, average grip strength, force rise time, force hold time, force-time integral, motion duration, motion amplitude, peak velocity, average velocity, and motion stability, as well as corresponding motion state data collected by a displacement sensor, and / or angle sensor, and / or inertial sensor, to form a behavioral state data parameter structure for characterizing the motion state of the target object.

[0067] In addition, for some target subjects who cannot complete obvious upper limb active movement tasks (such as obvious active grasping movements), the upper limb active movement tasks are marked as upper limb active movement attempt tasks. For example, active grasping tasks are marked as active grasping attempt tasks. At the same time, a grip strength meter or a high-sensitivity pressure sensor is used to record weak outputs. When no stable mechanical output is detected, the nominal task time window is determined based on the stimulus data. At the same time, this state is recorded in the trial matching degree and data reliability to avoid directly interpreting the failure to complete the task as a lack of central motor commands.

[0068] Step S200: Data preprocessing, data alignment, and task trial segmentation:

[0069] Data preprocessing and synchronization alignment are performed on the EEG signal data, stimulus data, and behavioral state data at each monitoring time point. The actual movement time is determined by combining the stimulus data and behavioral state data, and the task trials are segmented accordingly.

[0070] Data preprocessing: Channel quality checks are performed on the magnetoencephalogram (MEG) signal data to identify channels with excessive noise and missing data. Subsequently, preprocessing operations such as DC drift correction, target frequency band filtering, power frequency and harmonic suppression, and spatial background field correction are performed to suppress environmental noise. Furthermore, independent component decomposition and signal spatial projection can be used to remove eye movement, magnetocardiogram (MCC), and electromyography (EMG) artifacts, thus achieving physiological artifact removal. Behavioral state data undergoes preprocessing operations including, but not limited to, low-pass filtering, baseline correction, outlier correction, and unit unification to eliminate noise and correct the data.

[0071] Data alignment processing: Based on the shared trigger signal, or synchronization pulse, or device clock mapping relationship, the EEG signal data, stimulus data, and behavioral state data are time-synchronized and aligned. It should be noted that when the sampling rates of the EEG signal data, stimulus data, and behavioral state data are inconsistent, each data can be resampled to map them to a common synchronized time axis. Of course, the reversible sampling time mapping relationship can also be preserved.

[0072] Task trial segmentation: Combining stimulus data and behavioral state data, the actual movement time is determined. The actual movement time includes the actual movement start time, the movement maintenance phase, and the actual movement end time. At the same time, the task trial segmentation is divided into task time windows, including the pre-movement preparation window, the movement execution window, and the post-movement recovery window, based on the actual movement time, to ensure the consistency of task segmentation at different monitoring time points.

[0073] The actual motion start time is defined as the first moment when the mechanical state exceeds the start threshold relative to the baseline, and the actual motion end time is defined as the first moment when the mechanical state falls back below the end threshold and continues for a preset time.

[0074] Step S300: Mapping the brain magnetic signal data to a unified brain source space:

[0075] The magnetoencephalogram (MEG) signal data collected at each monitoring time point under each task is mapped to a unified brain source space based on an individualized head model or a standard head model, generating a spatiotemporal sequence of brain source activity corresponding to each task. Specifically, sensor location registration is performed on the MEG signal data collected at each monitoring time point, a forward model is calculated, and a source reconstruction method is used to map the MEG signal data to a unified brain source space, generating a spatiotemporal sequence of brain source activity corresponding to each task.

[0076] Specifically, during each monitoring session, the spatial geometric information of each sensor in the OPM-MEG system relative to the target subject's head, such as position and orientation, is recorded. Based on the anatomical coordinates of the individualized head model or a standard head model, sensor position registration is performed on the EEG signal data collected at each monitoring time point. The spatial geometric information of each sensor is acquired and recorded through 3D optical scanning, and / or binocular camera positioning, and / or laser scanning. Furthermore, when individual structural MRI images are available, signal data mapping can be directly performed based on the individualized head model. When individual structural MRI images are unavailable, the 3D head surface and anatomical landmarks are registered to the standard head model to complete signal data mapping. In both cases, it is necessary to maintain consistency in brain region division across time points so that longitudinal comparisons are performed in the internal brain-source space, rather than in the sensor channel space, which is susceptible to interference from external factors such as repeated wear.

[0077] Subsequently, the forward models of each sensor at different times were calculated, and the source reconstruction method was used to map the magnetoencephalogram (MEG) signal data to a unified brain source space, generating spatiotemporal sequences of brain source activity corresponding to each task. The aforementioned forward model is specifically represented as follows:

[0078]

[0079] In the formula, Indicates the first The monitoring time point, the first Task type at time Collected brain magnetic signal data, Indicates the first The forward operator at each monitoring time point is specifically calculated using the spatial geometric information of each sensor in the OPM-MEG system and the head model to establish a mapping between the brain source and the sensors. Indicates the first The monitoring time point, the first Task type at time The brain-derived activity is obtained by reverse engineering using methods such as beamforming, minimum norm estimation, and sparse source reconstruction. This indicates the noise term.

[0080] Step S400: Construct behavioral feature vectors and perform active-passive trial matching:

[0081] Based on the behavioral state data and the segmentation of task trials, behavioral feature vectors for upper limb active movement tasks and upper limb passive movement tasks are constructed. Active and passive trial matching is then performed on the two to obtain passive control feature vectors corresponding to the behavioral feature vectors of upper limb active movement task trials.

[0082] Specifically, behavioral feature vectors are constructed based on behavioral state data and task trial segmentation for upper limb active movement tasks and upper limb passive movement tasks. That is, for the first or second trial of an upper limb active movement task or upper limb passive movement task... Each trial is constructed based on the synchronously recorded behavioral state data and the task time window obtained by segmenting the task trials. Corresponding behavioral feature vectors .

[0083] Taking the grasping experiment implementation method mentioned above in this embodiment as an example, the corresponding behavioral feature vector is specifically represented as follows:

[0084]

[0085] In the formula, Indicates the first A behavioral feature vector of each active or passive grasping attempt. Indicates the first The peak grip force of an active grip test or the peak externally applied force of a passive grip test. Indicates the first The average force during the holding phase of each active or passive grasping attempt. Indicates the first The time it takes for the force to rise during each active or passive grasping attempt. Indicates the first The duration of force retention in each active or passive grasping attempt. Indicates the first The integral of force and time for each active or passive grasping attempt. Indicates the first Stability of the force curve for each active or passive gripping test.

[0086] Furthermore, the active-passive trial matching process in step S400 above specifically includes the following steps:

[0087] Step S410: Calculate the behavioral distance between the behavioral feature vectors of the upper limb active movement task trials and the behavioral feature vectors of the upper limb passive movement task trials.

[0088] That is: for each trial in the upper limb active movement task, calculate the behavioral distance between its behavioral feature vector and the behavioral feature vectors of each trial in the upper limb passive movement task, specifically expressed as:

[0089]

[0090] In the formula, Indicates the number of trials of active upper limb movement task Passive movement task of the upper limbs Behavioral distance between them This represents the number of features in the behavioral feature vector. Indicates the first The weight of each behavioral feature and , Indicates the first The normalized distance function of behavioral features, Indicates the number of trials of active upper limb movement task The first in the behavioral feature vector behavioral characteristics, Indicates the number of trials of the upper limb passive movement task The first in the behavioral feature vector A behavioral characteristic.

[0091] Step S420: Combine behavioral distance and preset matching threshold to filter and generate a candidate matching set that corresponds to the number of upper limb active movement task trials and consists of upper limb passive movement task trials;

[0092] Specifically, in this embodiment, behavioral distance is compared. Matching threshold Filter out all that meet the behavioral distance criteria. Less than the preset matching threshold Upper limb passive movement task trial Generate and test cases of active upper limb movement tasks Corresponding candidate matching set .

[0093] Step S430: If the candidate matching set contains only one upper limb passive movement task trial, that is, only one upper limb passive movement task trial meets the distance threshold screening condition, then a one-to-one pairing is established and the behavioral feature vector of the upper limb passive movement task trial is used as the passive control feature vector corresponding to the current upper limb active movement task trial.

[0094] If the candidate matching set contains multiple upper limb passive movement task trials, that is, multiple upper limb passive movement task trials simultaneously meet the distance threshold screening condition, then normalized weights are generated based on the behavioral distance between each upper limb passive movement task trial and the upper limb active movement task trial, and the behavioral feature vectors of multiple upper limb passive movement task trials are weighted and averaged to obtain the passive control feature vector corresponding to the upper limb active movement task trial.

[0095] Specifically, the calculation method for the corresponding passive control feature vector is as follows:

[0096]

[0097] In the formula, Indicates the first Upper limb active movement task test at each monitoring time point The corresponding passive control feature vector, Indicates the number of trials of active upper limb movement task The corresponding candidate matching set includes multiple trials of upper limb passive movement tasks. , Indicates the number of trials of the upper limb passive movement task Compared to the number of trials of the upper limb active movement task Normalized weights and satisfying Specifically, it can be calculated using methods such as the reciprocal of the distance, exponential decay, or distance sorting. Of course, it can also be verified and determined based on experience, such as through cross-validation or correlation analysis. Indicates the first Upper limb passive movement task test at each monitoring time point Behavioral feature vectors.

[0098] If the candidate matching set is empty, meaning no upper limb passive movement task trials meet the distance threshold screening condition, it indicates that the target object's movement ability is weak and cannot achieve the same amplitude or load as the upper limb passive movement task trials. In this case, a partial matching feature subset is generated based on the behavioral feature vector (such as movement direction, and / or start time, and / or duration, and / or measurable mechanical output, etc.). For example, in the specific implementation of the grasping experiment, the behavioral feature vector includes six behavioral features: peak grip force or peak externally applied force, average force during the holding phase, force rise time, force holding time, force time integral, and force curve stability. The partial matching feature subset can be formed by selecting some behavioral features from this subset. Then, partial screening and matching are performed based on the partial matching feature subset to generate a partial matching set. At the same time, the behavioral distance and the number of effective matches are included in the calculation of the data credibility index.

[0099] It is understandable that the above active-passive trial matching process can make the active-passive difference reflect more the active neural components than the differences in peripheral action conditions.

[0100] Step S500: Based on the spatiotemporal sequence of brain source activity for each task, construct a multi-brain region network covering the ipsilateral and contralateral sensorimotor brain regions (including M1, S1, SMA, PMC, and the parietal sensorimotor integration area) and extract the joint feature vector of each task from the multi-brain region network. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features.

[0101] Understandably, the ipsilateral hemisphere of the lesion refers to the hemisphere where the stroke lesion is located, while the contralateral hemisphere refers to the hemisphere opposite the stroke lesion. By combining analysis of multiple brain regions, network-level changes in motor planning, motor output, sensory input, and sensorimotor integration can be reflected simultaneously, avoiding the influence of compensatory activities on the response of a single brain region.

[0102] Specifically, in this embodiment, the first... Each monitoring time point corresponds to The source activity intensity characteristics of the task class are Time-frequency response characteristics are Activity timing characteristics are Brain region connectivity features Then the corresponding joint feature vector Specifically, it is expressed as follows:

[0103]

[0104] in, That is: task Corresponding to rest tasks Upper limb motor imagery task Active movement tasks of the upper limbs and upper limb passive movement tasks .

[0105] Source activity intensity characteristics include average source power, peak source power, spatial distribution within brain regions, and interhemispheric asymmetry. These can be calculated directly from the spatiotemporal sequence of brain source activity or based on changes in task relevance. Standardized extraction was performed to eliminate individual baseline differences.

[0106] Time-frequency response characteristics include Frequency band event-related desynchronization (ERD) Frequency band event-related desynchronization (ERD) and post-motion The rebound, specifically quantified by task-related power changes, affects brain regions. and frequency band Define task-related power changes Represented as:

[0107]

[0108] in, This indicates the brain region where the task phase is located. frequency band ,time power, Indicates the corresponding baseline power, when exist or When the frequency band is significantly less than 0, it indicates the ERD of the corresponding frequency band, after the motion ends. frequency band A value significantly greater than 0 indicates that after exercise... The rebound is used to obtain the time-frequency response characteristics.

[0109] The temporal characteristics of activity include the onset latency, peak latency, interbrain propagation delay, and whether brain activity leads or lags behind the actual grip force initiation time (actual movement time) in SMA, PMC, M1, and S1. These characteristics are specifically derived from the spatiotemporal sequence of brain-derived activity or the time series of task-related power changes in each brain region. For example, [the text abruptly ends here, likely due to an incomplete sentence or missing information]. The time when the activity first exceeds the preset threshold is defined as the activity initiation latency, the time when the maximum value is reached is defined as the peak latency, and the difference in latency between different brain regions is defined as the inter-brain propagation delay.

[0110] Brain region connectivity features include coherence, phase-locked values, and amplitude envelope correlation, which are calculated based on representative source time series of each brain region to reflect the strength of functional connectivity between brain regions.

[0111] The aforementioned joint feature vectors are standardized on baseline time windows, healthy side task data, or training sample statistics, and maintained according to brain regions, time windows, and feature types.

[0112] It is understandable that the joint feature vector formed by the combination of multiple types of features can simultaneously capture response amplitude, rhythm modulation, network propagation and brain region synergistic changes. Furthermore, the combination of multiple brain regions can distinguish between lesion-side recovery, contralateral compensation and cross-hemispheric reorganization. Therefore, compared to the existing single power index or single brain region response representing upper limb rehabilitation, this embodiment does not use a single power index or single brain region response to represent upper limb rehabilitation. Instead, it uses the combined features covering the bilateral sensorimotor networks as the subsequent differential decoupling input.

[0113] Step S600: Perform multi-task-state feature difference decoupling.

[0114] Based on the joint feature vectors corresponding to the resting task, upper limb motor imagery task, upper limb active motor task, and upper limb passive motor task, as well as the passive control feature vector corresponding to the upper limb active motor task, multi-task state feature difference decoupling processing is performed to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters.

[0115] The multi-task-state feature difference decoupling process specifically includes the following steps:

[0116] Step S610: Sensory afferent representation parameters are generated by differentially calculating the joint feature vector of the resting task and the joint feature vector of the upper limb passive movement task. These parameters are used to characterize changes in sensory afferent-related neural activity caused by external movement. Specifically, the sensory afferent representation parameters are obtained by differentially calculating the joint feature vector of the resting task and the joint feature vector of the upper limb passive movement task. Specifically, it is expressed as:

[0117]

[0118]

[0119] in, Indicates the first Sensory input differential features corresponding to each monitoring time point Indicates the first upper limb passive movement task at each monitoring time point The joint eigenvectors, Indicates the first Resting task at each monitoring time point The joint eigenvectors, Indicates the first Sensory input representation parameters corresponding to each monitoring time point This represents the sensory input mapping, which is specifically obtained through training or validation on multiple brain regions and multiple types of joint features.

[0120] It should be noted that the input representation parameters are perceptual. Specifically, it preserves components such as the S1 response intensity and latency in the sensorimotor brain region, S1 parietal lobe connectivity, and bilateral sensory network asymmetry, thereby avoiding the compression of sensory input into a single brain region value that lacks interpretability.

[0121] Step S620: Calculate and generate central motor instruction representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector.

[0122] Specifically, in this embodiment, the steps for calculating and obtaining the central motor command representation parameters are as follows:

[0123] Step S621: Calculate the differential features of the movement plan based on the joint feature vector of the upper limb motor imagery task and the joint feature vector of the resting task. :

[0124]

[0125] in, Indicates the first Upper limb motor imagery task at each monitoring time point The joint eigenvectors, Indicates the first Resting task at each monitoring time point The joint eigenvectors.

[0126] Step S622: Calculate the active participation differential features based on the joint feature vector of the upper limb active movement task and the passive control feature vector. :

[0127]

[0128] in, Indicates the first Upper limb active movement task at each monitoring time point The joint eigenvectors, This represents the passive control feature vector corresponding to the behavioral feature vector of each trial of the upper limb active movement task.

[0129] Step S623, based on the differential characteristics of the exercise plan and active participation in differential features Weighted calculation yields central motor command representation parameters :

[0130]

[0131] in, and The corresponding weight matrix is ​​configured according to the feature type and brain region, specifically determined by the healthy side task data, baseline data, or training samples.

[0132] It is understandable that the motor planning differential feature provides motor planning information when there is no obvious peripheral movement through motor imagery differential, while the active participation differential feature highlights the active participation component under similar sensory input conditions through active-passive differential. By combining the two, the bias caused by random fluctuations in single tasks and differences in peripheral load can be reduced.

[0133] Step S630: Generate sensorimotor integration representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector. These parameters are used to characterize the integration components generated by multi-brain region coordination and brain behavior coupling in the actual active response.

[0134] Specifically, the steps for calculating and obtaining the aforementioned sensorimotor integration representation parameters are as follows:

[0135] Step S631: Calculate the predicted active feature vector based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, and the passive control feature vector of the upper limb active motor task. :

[0136]

[0137] in, Indicates the first Upper limb motor imagery task at each monitoring time point The joint eigenvectors, The passive control feature vector represents the behavioral feature vector corresponding to each trial of the upper limb active movement task. Indicates the first Resting task at each monitoring time point The joint eigenvectors, This represents an active motion feature prediction model, which can be selected from any one of linear additive models, partial least squares regression, kernel ridge regression, Gaussian process regression, or shallow neural networks, and is trained using supervised learning. Its inputs retain brain regions, feature types, and time window indices, and its output has the same structure as the joint feature vector.

[0138] Step S632, based on the predicted active feature vector Joint feature vector of upper limb active motor task Calculated sensorimotor integration characterization parameters :

[0139]

[0140]

[0141] in, This represents sensorimotor integration mapping. Represents the joint feature vector of an upper limb active motor task. and predicting active feature vectors The residuals between them are used to characterize the remaining components in the actual active task features that cannot be explained by the motor plan, passive sensory input, and resting baseline.

[0142] Step S700: Generate longitudinal monitoring results and data reliability:

[0143] Longitudinal monitoring results are generated based on sensory afferent representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters at different monitoring time points. The longitudinal monitoring results include the change trajectory of sensory afferent representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters, as well as the affected-unaffected side control results and health reference standard scores.

[0144] Specifically, for the baseline time point and subsequent monitoring time points Sensory input representation parameters were obtained respectively. Central motor command representation parameters Sensorimotor integration representation parameters .

[0145] set up Indicates the sensory input representation parameters Central motor command representation parameters Sensorimotor integration representation parameters Any component in, i.e.:

[0146]

[0147] The longitudinal standard variation corresponding to baseline normalization is expressed as follows:

[0148]

[0149] in, Indicates the first Indicators at each monitoring time point The standardized change relative to the baseline. , They represent the first The index values ​​at each monitoring time point and baseline time point, , They represent the first The standard deviations of each monitoring time point and the baseline time point are obtained through trial resampling, cross-validation, and model uncertainty estimation.

[0150] Finally, the trajectory of change of sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters is obtained based on the changes of each longitudinal standard.

[0151] Furthermore, the results for the affected-healthy side comparison are specifically calculated based on the sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters of the brain regions corresponding to the affected and healthy limbs of the same patient. For example, the difference or ratio between the corresponding brain regions and corresponding feature components of the affected and healthy sides is calculated to eliminate individual differences and highlight side-specific changes. As for the health reference standard score, it is specifically calculated based on the pre-established distribution of sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters of the healthy control group. The mean and standard deviation of the patient's parameters are compared with the health reference distribution to calculate the standardized score, which is used to intuitively reflect the degree to which the patient deviates from the healthy state.

[0152] Furthermore, when the above results are consistent in direction and the quality conditions are met, the change trajectory results, the affected-healthy side control results, and the health reference standard scores are integrated to obtain comprehensive longitudinal monitoring result parameters.

[0153] Optionally, in some embodiments, indicators including but not limited to the proportion of effective sensors, the proportion of effective trials, head movement, ambient noise level, magnetoencephalography signal-to-noise ratio, source localization stability, active-passive trial matching degree, grip force signal integrity, and model prediction uncertainty are standardized into mass components. Simultaneously, indicators such as head motion, environmental noise level, and model prediction uncertainty are converted into inverse indicators with higher values ​​indicating higher quality. This is achieved by analyzing each quality component. Weighted summation yields data credibility :

[0154] ;

[0155] in, Indicates the first The weight of each quality component.

[0156] When data credibility When the data falls below a preset threshold, the system will output a prompt to re-collect data, or mark the current time point as low confidence. In this case, the abnormal changes at that time point will not be directly interpreted as rehabilitation-related neurological function changes.

[0157] Example 2:

[0158] In Embodiment 2 of the present invention, using Figure 3 This will be explained further. Additionally, the post-stroke upper limb sensorimotor function monitoring system based on multi-task OPM-MEG differential decoupling mentioned in this embodiment is used to execute the post-stroke upper limb sensorimotor function monitoring method based on multi-task OPM-MEG differential decoupling mentioned in Embodiment 1 above. For details of the method, please refer to the section of Embodiment 1 above; it will not be repeated here.

[0159] like Figure 3 As shown, the present invention relates to a post-stroke upper limb sensory and motor function monitoring system based on multi-task OPM-MEG differential decoupling, which includes a data acquisition module, a data processing module, a source space mapping module, a matching processing module, a feature extraction module, a differential decoupling module, and a longitudinal monitoring module.

[0160] The data processing module is connected to the data acquisition module, the source space mapping module and the matching processing module are both connected to the data processing module, the feature extraction module is connected to the source space mapping module, the matching processing module and the feature extraction module are both connected to the differential decoupling module, and the longitudinal monitoring module is connected to the differential decoupling module.

[0161] The data acquisition module is used to collect the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target object when performing resting tasks, upper limb motor imagery tasks, upper limb active movement tasks, and upper limb passive movement tasks at at least two monitoring time points.

[0162] The data processing module is used to preprocess and synchronize the brain magnetic signal data, stimulus data, and behavioral state data at each monitoring time point. It combines the stimulus data and behavioral state data to determine the actual movement time and thereby divide the task trials into segments.

[0163] The source space mapping module is used to map the EEG signal data collected at each monitoring time point under each task to a unified brain source space based on an individualized head model or a standard head model, and generate a spatiotemporal sequence of brain source activity corresponding to each task.

[0164] The matching processing module is used to construct behavioral feature vectors for upper limb active movement tasks and upper limb passive movement tasks based on behavioral state data and task trial segmentation, and to perform active-passive trial matching processing on the two to obtain a passive control feature vector corresponding to the behavioral feature vector of the upper limb active movement task trial.

[0165] The feature extraction module is used to construct a multi-brain region network covering the ipsilateral and contralateral sensorimotor brain regions of the lesion based on the spatiotemporal sequence of brain source activity for each task, and to extract the joint feature vector for each task. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features.

[0166] The differential decoupling module is used to perform multi-task state feature differential decoupling processing to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters based on the joint feature vectors corresponding to resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks, as well as the passive control feature vector corresponding to upper limb active motor tasks.

[0167] The longitudinal monitoring module is used to generate longitudinal monitoring results based on sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters at different monitoring time points.

[0168] It should be noted that the above modules can be deployed on a data processing computer, a server, or a clinical terminal, and transmit data through a database, and / or shared storage, and / or a network interface.

[0169] In summary, the method and system for monitoring upper limb sensory and motor function after stroke based on multi-task OPM-MEG differential decoupling provided by this invention reduces the impact of factors such as different sensor positions, different levels of task completion, and differences in patient passive feedback in cross-treatment time comparison analysis by differential decoupling of multi-task data, thereby improving the accuracy and reliability of longitudinal monitoring results.

[0170] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. All equivalent changes and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling, characterized in that, Includes the following steps: Step S100: Collect and obtain the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target object when performing resting task, upper limb motor imagery task, upper limb active motor task, and upper limb passive motor task at at least two monitoring time points respectively; Step S200: Perform data preprocessing and synchronization alignment processing on the brain magnetic signal data, stimulus data, and behavioral state data at each monitoring time point, and combine the stimulus data and behavioral state data to determine the actual movement time and thereby divide the task trials into segments. Step S300: Map the brain magnetic signal data collected at each monitoring time point under each task to a unified brain source space based on an individualized head model or a standard head model, and generate a brain source activity spatiotemporal sequence corresponding to each task. Step S400: Construct behavioral feature vectors for upper limb active movement task and upper limb passive movement task trials based on behavioral state data and task trial segmentation, and perform active-passive trial matching processing on the two to obtain a passive control feature vector corresponding to the behavioral feature vector of the upper limb active movement task trial. Step S500: Based on the spatiotemporal sequence of brain source activity for each task, construct a multi-brain region network covering the sensorimotor brain regions on the same side and opposite side of the lesion, and extract the joint feature vector for each task. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features. Step S600: Based on the joint feature vectors corresponding to the resting task, upper limb motor imagery task, upper limb active motor task, and upper limb passive motor task, and the passive control feature vector corresponding to the upper limb active motor task, perform multi-task state feature difference decoupling processing to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters. Step S700: Generate longitudinal monitoring results based on the sensory input representation parameters, the central motor command representation parameters, and the sensorimotor integration representation parameters at different monitoring time points.

2. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 1, characterized in that, The active-passive trial matching process includes the following steps: Step S410: Calculate the behavioral distance between the behavioral feature vectors of the upper limb active movement task trials and the behavioral feature vectors of the upper limb passive movement task trials. Step S420: Combine the behavioral distance and the preset matching threshold to filter and generate a candidate matching set that corresponds to the upper limb active movement task trials and consists of upper limb passive movement task trials; Step S430: If the candidate matching set contains only one upper limb passive movement task trial, then establish a one-to-one pairing and use the behavioral feature vector of the upper limb passive movement task trial as the passive control feature vector corresponding to the current upper limb active movement task trial. If the candidate matching set contains multiple upper limb passive movement task trials, then normalized weights are generated based on the behavioral distance between each upper limb passive movement task trial and the upper limb active movement task trial, and the behavioral feature vectors of multiple upper limb passive movement task trials are weighted and averaged to obtain the passive control feature vector corresponding to the upper limb active movement task trial.

3. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 1, characterized in that, The multi-task-state feature difference decoupling process includes the following steps: Step S610: Generate the sensory input representation parameters by differential calculation based on the joint feature vector of the resting task and the joint feature vector of the upper limb passive movement task; Step S620: Calculate and generate the central motor instruction representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector. Step S630: Calculate and generate the sensorimotor integration representation parameters based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, the joint feature vector of the upper limb active motor task, and the passive control feature vector.

4. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 3, characterized in that, The sensory input representation parameters are calculated and obtained as follows: ； ； in, Indicates the first Sensory input differential features corresponding to each monitoring time point Indicates the first upper limb passive movement task at each monitoring time point The joint eigenvectors, Indicates the first Resting task at each monitoring time point The joint eigenvectors, Indicates the first Sensory input representation parameters corresponding to each monitoring time point This indicates that the input mapping is perceptual.

5. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 3, characterized in that, The calculation and acquisition of the central motion command representation parameters include the following steps: Step S621: Calculate the differential features of the movement plan based on the joint feature vector of the upper limb motor imagery task and the joint feature vector of the resting task; Step S622: Calculate the active participation differential features based on the joint feature vector of the upper limb active movement task and the passive control feature vector; Step S623: The central motor command representation parameters are obtained by weighted calculation based on the differential features of the movement plan and the differential features of active participation.

6. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 5, characterized in that, The calculation and acquisition method for the central motion command representation parameters is as follows: ； in, Indicates the first The central motor command characterization parameters corresponding to each monitoring time point and These represent the corresponding weight matrices. Indicates the first Differential features of exercise plans corresponding to each monitoring time point express Active participation differential features corresponding to each monitoring time point.

7. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 3, characterized in that: The calculation and acquisition of the sensorimotor integration representation parameters include the following steps: Step S631: Calculate the predicted active feature vector based on the joint feature vector of the upper limb motor imagery task, the joint feature vector of the resting task, and the passive control feature vector of the upper limb active motor task. Step S632: Calculate the sensorimotor integration representation parameters based on the predicted active feature vector and the joint feature vector of the upper limb active movement task.

8. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 7, characterized in that, The sensorimotor integration representation parameters are calculated and obtained as follows: ； ； in, Indicates the first Sensorimotor integration characterization parameters corresponding to each monitoring time point This represents sensorimotor integration mapping. Indicates the first Joint feature vector of upper limb active movement task at each monitoring time point and predicting active feature vectors The residuals between them.

9. The method for monitoring upper limb sensory and motor function after stroke based on multi-task state OPM-MEG differential decoupling according to claim 1, characterized in that: The longitudinal monitoring results include the trajectory of changes in sensory afferent representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters, as well as the affected-unaffected side comparison results and health reference standard scores.

10. A post-stroke upper limb sensorimotor function monitoring system based on multi-task OPM-MEG differential decoupling, used to execute the post-stroke upper limb sensorimotor function monitoring method based on multi-task OPM-MEG differential decoupling as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect the brain magnetic signal data, stimulus data, and behavioral state data generated synchronously by the target object when performing resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks at at least two monitoring time points. The data processing module is used to preprocess and synchronize the brain magnetic signal data, stimulus data, and behavioral state data at each monitoring time point, and to determine the actual movement time by combining the stimulus data and behavioral state data, and thereby divide the task trials into segments. The source space mapping module is used to map the magnetoencephalogram (MEG) signal data collected at each monitoring time point under each task to a unified brain source space based on an individualized head model or a standard head model, and generate a spatiotemporal sequence of brain source activity corresponding to each task. The matching processing module is used to construct behavioral feature vectors for upper limb active movement tasks and upper limb passive movement tasks based on behavioral state data and task trial segmentation, and to perform active and passive trial matching processing on the two to obtain a passive control feature vector corresponding to the behavioral feature vector of the upper limb active movement task trial. The feature extraction module is used to construct a multi-brain region network covering the ipsilateral and contralateral sensorimotor brain regions of the lesion based on the spatiotemporal sequence of brain source activity for each task, and to extract the joint feature vector for each task. The joint feature vector includes source activity intensity features, time-frequency response features, activity temporal features, and brain region connectivity features. The differential decoupling module is used to perform multi-task state feature differential decoupling processing to obtain sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters based on the joint feature vectors corresponding to resting tasks, upper limb motor imagery tasks, upper limb active motor tasks, and upper limb passive motor tasks, as well as the passive control feature vector corresponding to upper limb active motor tasks. The longitudinal monitoring module is used to generate longitudinal monitoring results based on sensory input representation parameters, central motor command representation parameters, and sensorimotor integration representation parameters at different monitoring time points.

Images