Patient rehabilitation tracking system, rehabilitation training regulation methods, equipment and media based on active perception and environmental adaptation

Abstract

This application provides a patient rehabilitation tracking system, a rehabilitation training control method, device, and medium based on active perception and environmental adaptation. It determines the coordinated movement state of the neuromuscular system based on the multimodal physiological data stream of the target patient's rehabilitation training, and then performs intention-muscle force coordination identification based on the coordinated movement state to obtain the target patient's movement intention and muscle force contribution distribution. It also determines a dynamic risk map of the target patient during rehabilitation training; performs variable-domain fuzzy reasoning on the movement intention, muscle force contribution distribution, and dynamic risk map to obtain rehabilitation adjustment instructions matching the target patient's rehabilitation state; and continuously tracks the target patient's rehabilitation progress and controls a rehabilitation training robot to perform adaptive assistance compensation based on the rehabilitation adjustment instructions. Using the solution of this application, the patient's rehabilitation training can be decoded for movement intention and analyzed for muscle force contribution, and rehabilitation training can be controlled in conjunction with dynamic environmental risk.

Description

Technical Field

[0001] This application relates to the field of treatment improvement technology, and more specifically, to a patient rehabilitation tracking system, a rehabilitation training regulation method, device and medium based on active perception and environmental adaptation. Background Technology

[0002] With the increasing aging of the population, the number of patients with motor dysfunction caused by diseases such as stroke and nerve injury has increased significantly. Rehabilitation training has become a core means to improve patients' limb motor ability and promote the remodeling of nerve function. With the widespread application of rehabilitation robot technology in neuro-musculoskeletal rehabilitation, the current rehabilitation field is upgrading from traditional passive assisted training to active rehabilitation training. How to achieve accurate tracking and personalized intervention of patients' rehabilitation process has become a research hotspot.

[0003] However, existing rehabilitation tracking methods have the following technical limitations: First, at the data acquisition level, they mostly use a single sensor, such as using only an inertial measurement unit to collect limb movement data or only a surface electromyography (SEMG) sensor to collect electromyographic signals. They lack spatiotemporal fusion processing of limb movement data and electromyographic physiological signals, making it impossible to accurately characterize the coordinated movement state of the nervous and musculoskeletal systems. Second, at the identification level, they can only infer movement intentions from movement trajectories, failing to accurately quantify the distribution of muscle force contributions from different muscle groups. Finally, at the environmental adaptation level, they do not consider the dynamic spatial domain of obstacles in the rehabilitation environment, relying solely on fixed safe distances to assess risk. This can easily lead to collision hazards due to a lack of understanding of human-computer interaction and obstacle risk assessment. Therefore, how to decode movement intentions and analyze muscle force contributions in patients' rehabilitation training, and how to combine this with dynamic environmental risk control for rehabilitation training, has become a challenge for the industry. Summary of the Invention

[0004] This application provides a patient rehabilitation tracking system, a rehabilitation training regulation method, device, and medium based on active perception and environmental adaptation, which can decode the patient's motor intention and analyze muscle strength contribution during rehabilitation training, and regulate rehabilitation training in combination with dynamic environmental risks.

[0005] Firstly, this application provides a rehabilitation training regulation method based on active perception and environmental adaptation, used in a patient rehabilitation tracking system for adaptive rehabilitation training regulation of target patients. The method includes:

[0006] Simultaneously collect limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, and then obtain a multimodal physiological data stream for tracking the rehabilitation training of the target patient;

[0007] Based on the multimodal physiological data stream, the coordinated movement state of the neuromusculoskeletal system during the rehabilitation training of the target patient is determined, and then the intention-muscle force coordination identification is performed based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient based on the decoding of neurophysiological signals during rehabilitation training.

[0008] A dynamic risk map of the target patient during rehabilitation training is determined by the interaction between the target patient and the rehabilitation training robot in the rehabilitation training environment and the spatial domain occupied by obstacles.

[0009] The exercise intention, the muscle strength contribution distribution and the dynamic risk map are subjected to variable domain fuzzy reasoning, and then the rehabilitation adjustment instructions that match the rehabilitation status of the target patient are determined based on the reasoning results.

[0010] The rehabilitation progress of the target patient is continuously tracked, and the rehabilitation training robot is controlled to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

[0011] In one implementation of the first aspect, the simultaneous acquisition of limb movement data and electromyographic signals of the target patient during rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the target patient's rehabilitation training, specifically includes:

[0012] The inertial measurement unit collects limb movement data of the target patient when performing rehabilitation training tasks;

[0013] Electromyographic signals of target patients during rehabilitation training tasks were acquired using a surface electromyography sensor array.

[0014] The limb movement data and the electromyographic signals are time-stamped to form a multimodal physiological data stream for tracking the rehabilitation training of the target patient.

[0015] In another implementation of the first aspect, determining the coordinated motor state of the neuromuscular system during rehabilitation training of the target patient based on the multimodal physiological data stream specifically includes:

[0016] Logarithmic time-domain features of electromyographic physiological signals in the multimodal physiological data stream are extracted, and then an electromyographic feature matrix is ​​constructed.

[0017] Extract joint angle information of the target patient during limb movement from the multimodal physiological data stream;

[0018] By spatiotemporally coupling the electromyographic feature matrix with the joint angle information, the coordinated movement state of the neuromuscular system during the rehabilitation training of the target patient can be obtained.

[0019] In another implementation of the first aspect, the intention-muscle force coordination identification based on the aforementioned coordinated movement state, to obtain the distribution of movement intention and muscle force contribution based on neurophysiological signal decoding during rehabilitation training of the target patient, specifically includes:

[0020] By combining the target patient's electroencephalogram (EEG) signals, the coordinated motor state is decoded hierarchically to obtain the motor coding vector of the target patient's motor cortex;

[0021] The motion encoding vector is used to determine the target patient's motion intention based on the decoding of neurophysiological signals during rehabilitation training;

[0022] Based on the electromyography-torque mapping relationship of the target patient, the force contribution of each muscle group in the target patient is determined according to the synergistic motion state;

[0023] The distribution of muscle force contribution during rehabilitation training for the target patient is generated based on the force contribution of all muscle groups.

[0024] In another implementation of the first aspect, determining the dynamic risk map of the target patient during rehabilitation training by using the interaction state between the target patient and the rehabilitation training robot in the rehabilitation training environment and the spatial domain occupied by obstacles specifically includes:

[0025] Real-time monitoring of the spatial location of obstacles in the rehabilitation training environment to obtain the motion status of the obstacles;

[0026] The spatial domain occupied by the obstacle is determined based on the motion state;

[0027] Capture the relative pose relationship between the target patient and the rehabilitation training robot in the rehabilitation training environment, and then determine the interaction state between the target patient and the rehabilitation training robot;

[0028] Multi-source perception fusion is performed on the spatial domain and the interaction state to establish a dynamic risk map of the rehabilitation training environment.

[0029] In another implementation of the first aspect, performing variable-domain fuzzy reasoning on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, and then determining rehabilitation adjustment instructions matching the rehabilitation status of the target patient based on the reasoning results, specifically includes:

[0030] Construct a multidimensional rehabilitation decision space with fuzzy reasoning between clarity of motor intent, muscle strength coordination efficiency, and environmental risk level;

[0031] Through the adaptive domain adjustment mechanism of the rehabilitation stage in the multidimensional rehabilitation decision space, combined with the real-time movement deviation of the target patient, the variable domain scaling factor of multimodal collaborative reasoning is determined according to the movement intention, the muscle strength contribution distribution and the dynamic risk map.

[0032] Based on the variable universe scaling factor, the universe of discourse of the fuzzy inference is corrected, and then the intensity correction coefficient and the auxiliary force adjustment coefficient of the rehabilitation training are obtained.

[0033] Based on the intensity correction coefficient and the auxiliary force adjustment coefficient, a rehabilitation adjustment instruction matching the rehabilitation status of the target patient is generated.

[0034] In another implementation of the first aspect, continuously tracking the rehabilitation progress of the target patient and controlling the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions specifically includes:

[0035] Based on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, a dynamic rehabilitation efficacy index is constructed to continuously track the rehabilitation progress of the target patient.

[0036] Based on the rehabilitation adjustment instructions and the dynamic rehabilitation efficacy indicators, the rehabilitation training robot is controlled to provide impedance-adaptive dynamic assistance compensation to the target patient.

[0037] Secondly, this application provides a patient rehabilitation tracking system, which includes a rehabilitation training control unit, the rehabilitation training control unit comprising:

[0038] The acquisition module is used to synchronously acquire limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the rehabilitation training of the target patient;

[0039] The processing module is used to determine the coordinated movement state of the neuromusculoskeletal system during the rehabilitation training of the target patient based on the multimodal physiological data stream, and then perform intention-muscle force coordination identification based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient based on the decoding of neurophysiological signals during rehabilitation training.

[0040] The processing module is used to determine a dynamic risk map of the target patient during rehabilitation training by using the interaction state between the target patient and the rehabilitation training robot and the spatial domain occupied by obstacles in the rehabilitation training environment.

[0041] The processing module is used to perform variable domain fuzzy reasoning on the movement intention, the muscle strength contribution distribution and the dynamic risk map, and then determine rehabilitation adjustment instructions that match the rehabilitation status of the target patient based on the reasoning results.

[0042] The execution module is used to continuously track the rehabilitation progress of the target patient and control the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

[0043] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described rehabilitation training regulation method based on active perception and environmental adaptation.

[0044] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described rehabilitation training regulation method based on active perception and environmental adaptation.

[0045] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0046] The patient rehabilitation tracking system, rehabilitation training control method, equipment, and medium provided in this application first synchronously collect limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the target patient's rehabilitation training; determine the coordinated movement state of the neuromuscular system in the target patient's rehabilitation training based on the multimodal physiological data stream, and then perform intention-muscle force coordination identification based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient during rehabilitation training based on the decoding of neurophysiological signals; determine the dynamic risk map of the target patient during rehabilitation training by the interaction state between the target patient and the rehabilitation training robot and the spatial domain occupied by obstacles in the rehabilitation training environment; perform variable domain fuzzy reasoning on the movement intention, the muscle force contribution distribution, and the dynamic risk map, and then determine rehabilitation adjustment instructions matching the target patient's rehabilitation state based on the reasoning results; continuously track the target patient's rehabilitation progress and control the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

[0047] Therefore, this application continuously tracks the rehabilitation progress of the target patient and controls the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions. First, determining the movement intention yields a set of movement instructions describing the rehabilitation movement trajectory that the target patient expects to perform. The determination of movement intention can be achieved by parsing the movement coding vector of the target patient's motor cortex through multimodal physiological data stream, thereby accurately capturing the central nervous system's planning instructions for rehabilitation movements. This makes movement intention a quantitative indicator reflecting the patient's true neural drive needs, avoiding the subjectivity and bias of traditional trajectory-based intention deduction, and providing a benchmark for active rehabilitation training based on the patient's neural will. Simultaneously, by accurately identifying the patient's active movements... The intention to assist ensures that the adaptive assistance compensation of the rehabilitation robot only provides assistance when the patient's intention is clear but muscle strength is insufficient, rather than being passively driven throughout the process. This strengthens the connection between the central nervous system and muscles and enhances the patient's sense of participation and confidence in training. Then, determining the distribution of muscle strength contribution yields a synergistic pattern atlas reflecting the activation sequence and intensity relationship of various muscles during limb movement in the target patient during rehabilitation training. The determination of muscle strength contribution distribution can be visually presented by extracting the activation sequence and intensity ratio of dominant and accessory muscle groups, thus providing a direct representation of the synergistic activation pattern of the neuromuscular system. Furthermore, combining this with the analysis of the synergistic movement state of the neuromuscular system allows for further tracing of the compensatory root causes (such as insufficient neural drive or muscle dysfunction). (Deficiency) and provides quantitative basis for subsequent adjustment of training strategies, avoiding the blind spot of traditional methods that only look at the movement trajectory and ignore the assessment of muscle group coordination. This enables dynamic adjustment of assistance according to muscle group exertion gaps, allowing the rehabilitation robot's assistance compensation to accurately match the real-time functional level of each muscle group in the patient. This ensures both the completion of the target movement and strengthens the patient's awareness of active exertion, which is in line with the core goal of active rehabilitation training to promote the remodeling of neural function. Finally, determining the dynamic risk map can obtain an overall safety distribution map that presents the real-time risk level of different areas in the rehabilitation training environment. The determination of the dynamic risk map, through multi-source perception fusion, can distinguish the static spatial domain of static obstacles (such as training racks) and dynamic Obstacles (such as medical personnel and mobile equipment) include a dynamic spatial domain predicting their location within a specified future timeframe, thus accurately defining the risk impact range of different obstacles. On the other hand, the human-computer interaction state can be transformed into quantitative risk indicators and integrated into the global environmental assessment, forming a global safety view that presents risk levels in gridded units. This solves the problem in existing solutions where local risk assessments cannot reflect the overall safety status, providing a complete environmental safety basis for subsequent rehabilitation adjustment instructions and preventing safety accidents caused by the omission of local risks. In summary, based on the above solution, rehabilitation training for patients can be modified by decoding motor intentions and analyzing muscle contribution, combined with dynamic environmental risks. Attached Figure Description

[0048] Figure 1This is an exemplary flowchart of a rehabilitation training regulation method based on active perception and environmental adaptation, as shown in some embodiments of this application.

[0049] Figure 2 This is an operation flowchart illustrating the determination of the cooperative motion state according to some embodiments of this application;

[0050] Figure 3 This is an exemplary flowchart illustrating the determination of a dynamic risk map according to some embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of a rehabilitation training control unit according to some embodiments of this application;

[0052] Figure 5 This is an internal structural diagram of a computer device that implements a rehabilitation training regulation method based on active perception and environmental adaptation, according to some embodiments of this application. Detailed Implementation

[0053] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] refer to Figure 1 The figure is an exemplary flowchart of a rehabilitation training regulation method based on active perception and environmental adaptation, according to some embodiments of this application. This rehabilitation training regulation method based on active perception and environmental adaptation mainly includes the following steps:

[0055] In step 101, limb movement data and electromyographic signals of the target patient are collected simultaneously when performing rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the rehabilitation training of the target patient.

[0056] In some embodiments, the simultaneous acquisition of limb movement data and electromyographic signals of the target patient during rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the target patient's rehabilitation training, can be achieved through the following steps:

[0057] The inertial measurement unit collects limb movement data of the target patient when performing rehabilitation training tasks;

[0058] Electromyographic signals of target patients during rehabilitation training tasks were acquired using a surface electromyography sensor array.

[0059] The limb movement data and the electromyographic signals are time-stamped to form a multimodal physiological data stream for tracking the rehabilitation training of the target patient.

[0060] In specific implementation, the acquisition of limb motion data of the target patient during rehabilitation training tasks through inertial measurement units can be achieved in the following way: Inertial measurement unit sensors (such as the MPU9250) integrating a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer can be used to acquire limb motion data of the target patient during rehabilitation training tasks at a sampling frequency of 100Hz. Specifically, multiple inertial measurement unit sensors can be fixed to the target patient's upper arm, forearm, thigh, and lower leg respectively for data acquisition. A complementary filtering algorithm based on quaternions is used to fuse the accelerometer and gyroscope data, and then attitude calculation based on gradient descent is used to further fuse the magnetometer data. Gyroscope drift errors are corrected, and the set of three-dimensional attitude data of each limb segment in the global coordinate system is output as the limb motion data of the target patient during rehabilitation training tasks. The limb motion data is a dynamic information set representing the spatial position, joint angles, and movement trajectory of the limbs during rehabilitation exercises. This limb motion data can provide objective and quantifiable motor function indicators for rehabilitation assessment, enabling precise tracking of the standardization and fluency of rehabilitation training movements.

[0061] In practice, the acquisition of electromyographic signals of the target patient during rehabilitation training tasks using a surface electromyography (SEMG) sensor array can be achieved in the following manner: First, multiple differential SEMG sensors are arranged according to the anatomical location of the target muscles in the main muscle groups of the affected limb of the target patient, such as the anterior deltoid, biceps brachii, and triceps brachii, and raw EEMG signals are acquired at a sampling frequency of 1000 Hz. Then, the acquired raw EEMG signals are bandpass filtered at 500 Hz to remove motion artifacts and high-frequency noise. The set of all filtered raw EEMG signals is then used as the electromyographic signal of the target patient during rehabilitation training tasks. The electromyographic signal is a bioelectrical signal that characterizes the activation state and contraction level of the neuromuscular system of the target patient during rehabilitation training. This electromyographic signal can reflect the target patient's active movement intention and muscle exertion level, providing a key basis for distinguishing between active movement and passive assistance and assessing neural drive ability.

[0062] It should be noted that, in this application, the multimodal physiological data stream is a comprehensive data sequence with a unified spatiotemporal reference characterizing changes in limb spatial position and activation state and contraction level of the neuromuscular system during the rehabilitation training of the target patient. This multimodal physiological data stream breaks down data silos between different physiological information sources, providing a complete and synchronous multi-source data foundation for subsequent intention-muscle coordination identification and whole-body coordinated movement state analysis. Specifically, aligning the limb movement data and the electromyographic physiological signals with timestamps to form a multimodal physiological data stream for tracking the rehabilitation training of the target patient can be achieved in the following way: First, it can be... The limb movement data and electromyographic (EMG) physiological signals are resampled. The high sampling rate EMG data is downsampled to the same sampling rate as the limb movement data after being filtered by anti-aliasing. Then, the time offset corresponding to the maximum cross-correlation coefficient of the limb movement data and EMG physiological signals in the time domain is calculated using a time-series registration algorithm based on cross-correlation function. The remaining micro-time-series deviation is compensated for to perform time-series registration. The time-series registered limb movement data and EMG physiological signals are then combined according to the time sequence. Each time point contains the corresponding joint movement parameters and muscle activation parameters to construct a multimodal physiological data stream with a unified timestamp.

[0063] In step 102, the coordinated movement state of the neuromusculoskeletal system during the rehabilitation training of the target patient is determined based on the multimodal physiological data stream. Then, based on the coordinated movement state, intention-muscle force coordination identification is performed to obtain the movement intention and muscle force contribution distribution of the target patient during rehabilitation training based on the decoding of neurophysiological signals.

[0064] In some embodiments, reference Figure 2 The figure is a flowchart illustrating the operation of determining the synergistic motion state according to some embodiments of this application. In this application, determining the synergistic motion state of the neuromuscular system in the rehabilitation training of a target patient based on the multimodal physiological data stream can be achieved by the following steps:

[0065] Logarithmic time-domain features of electromyographic physiological signals in the multimodal physiological data stream are extracted, and then an electromyographic feature matrix is ​​constructed.

[0066] Extract joint angle information of the target patient during limb movement from the multimodal physiological data stream;

[0067] By spatiotemporally coupling the electromyographic feature matrix with the joint angle information, the coordinated movement state of the neuromuscular system during the rehabilitation training of the target patient can be obtained.

[0068] In specific implementation, the extraction of logarithmic time-domain features of electromyographic (EMG) physiological signals from the multimodal physiological data stream, and the subsequent construction of the EMG feature matrix, can be achieved in the following manner: First, the time sequence of EMG physiological signals can be extracted from the multimodal physiological data stream, and the time sequence can be segmented using the sliding time window method. The window length is set to 200ms, and within each time window, the time-domain features are calculated, including: the absolute value of the logarithmic mean, which is obtained by taking the absolute value of the time sequence, calculating the arithmetic mean, and then taking the natural logarithm; and the root mean square of the logarithmic mean, which is obtained by taking the square root of the square of the time sequence, calculating the average, and then taking the natural logarithm. The logarithmic waveform length is obtained by taking the natural logarithm of the sum of the absolute values ​​of the first-order differences of the time series; the logarithmic standard deviation is obtained by taking the natural logarithm of the standard deviation of the time series; finally, these time-domain features are expanded according to the feature dimensions, forming feature vectors within each time window, and stacked along the time series to construct an electromyographic feature matrix; wherein, the electromyographic feature matrix is ​​a multi-dimensional feature set that quantifies the muscle activation intensity and pattern of the target patient during rehabilitation training. This electromyographic feature matrix can transform the original chaotic electrical signals into a stable and computable data structure, providing reliable input features for subsequent accurate identification of neural control patterns.

[0069] In specific implementation, extracting the joint angle information of the target patient during limb movement from the multimodal physiological data stream can be achieved in the following way: the temporal sequence of limb movement data can be extracted from the multimodal physiological data stream, and the angle change curves of each key joint in the motion plane during limb movement of the target patient's affected limb can be calculated based on the joint point positions of the human contour skeleton using an existing human skeletal model (such as the lower limb biomechanical model based on Opensim). The set of all key joint angle change curves is then used as the joint angle information of the target patient during limb movement. Among them, key joints may include hip, knee, ankle, or shoulder, elbow, etc. The joint angle information is a spatiotemporal parameter describing the range and trajectory of joint movement during limb movement of the target patient. This joint angle information can objectively quantify the spatial movement of the limb and is an important basis for assessing whether the target patient's movement pattern is normal and whether there is compensation. At the same time, it can establish a kinematic benchmark for analyzing the standardization of movement execution and its correlation with muscle activation.

[0070] In specific implementation, the electromyographic feature matrix is ​​spatiotemporally coupled with the joint angle information to obtain the neuromusculoskeletal coordinated movement state during the rehabilitation training of the target patient. This can be achieved in the following way: a non-negative matrix factorization algorithm can be used to decompose the electromyographic feature matrix into multiple coordinated vectors representing muscle coordinated activation modes and activation coefficients controlling the temporal activation of these modes. A hybrid similarity measurement algorithm (such as a multi-index fusion method combining mutual information and dynamic time warping) is used to analyze the temporal correlation between these coordinated vectors and the angle change curves in the joint angle information, thereby decoupling the coordinated movement state that characterizes the motor control strategy during limb movement, i.e., how the central nervous system coordinates multiple muscles to work together to produce a specified joint movement. The coordinated movement state reflects the control mechanism of the central nervous system in the target patient's limb movement during rehabilitation training, which coordinates multiple muscles to work together to produce the target joint movement through a combination of specified muscle coordinated units. This coordinated movement state can determine whether the target patient's neural strategy for movement is normal, providing a key basis for identifying abnormal compensatory movements and accurately locating neurological deficits.

[0071] In some embodiments, intention-muscle strength coordination identification based on the coordinated movement state to obtain the movement intention and muscle strength contribution distribution of the target patient during rehabilitation training based on neurophysiological signal decoding can be achieved through the following steps:

[0072] By combining the target patient's electroencephalogram (EEG) signals, the coordinated motor state is decoded hierarchically to obtain the motor coding vector of the target patient's motor cortex;

[0073] The motion encoding vector is used to determine the target patient's motion intention based on the decoding of neurophysiological signals during rehabilitation training;

[0074] Based on the electromyography-torque mapping relationship of the target patient, the force contribution of each muscle group in the target patient is determined according to the synergistic motion state;

[0075] The distribution of muscle force contribution during rehabilitation training for the target patient is generated based on the force contribution of all muscle groups.

[0076] In specific implementation, the coordinated movement state is decoded hierarchically by combining the target patient's EEG signals to obtain the motion coding vector of the target patient's motor cortex. This can be achieved in the following way: First, the scalp EEG signals of the target patient during rehabilitation training tasks can be synchronously acquired using a multi-channel EEG acquisition device at a sampling frequency of 1000Hz. Independent component analysis (ICA) is then used to preprocess the raw EEG signals to remove artifacts such as electrooculography (EOG) and electromyography (EMG). Next, wavelet packet transform is used to extract event-related desynchronization features and event-related synchronization features in the motion-related frequency bands from the preprocessed raw EEG signals. These event-related desynchronization features and event-related synchronization features are then concatenated to construct an EEG feature vector with spatiotemporal characteristics. Finally, a deep convolutional neural network with an encoder-decoder structure is constructed based on this EEG feature vector. The encoder part extracts spatiotemporal pattern features from the EEG feature vector and EMG features through multi-layer convolution operations, while the decoder part uses deconvolution layers and full convolutions... The connection layer maps high-level abstract features to the motion parameter space. The deep convolutional neural network is trained using a mean squared error loss function and synchronously acquired limb movement trajectories as supervision signals. The network parameters are optimized through an error backpropagation algorithm, ultimately outputting a multi-dimensional motion encoding vector containing movement direction, velocity, and amplitude. Finally, through hierarchical feature selection and attention mechanisms, electromyographic features related to motor cortex control are screened in the coordinated movement state. The electromyographic and electroencephalographic feature vectors are input into the trained deep convolutional neural network, which decodes the motion encoding vector of the target patient's motor cortex. The motion encoding vector refers to the set of high-dimensional neural instructions for the motor cortex to plan limb movements when the target patient performs limb movements during rehabilitation training. This motion encoding vector realizes the inverse mapping from surface muscle electromyographic signals to central nervous system instructions, providing the most direct neurophysiological basis for accurately identifying the patient's true movement intentions.

[0077] In specific implementation, determining the target patient's motor intention based on neurophysiological signal decoding during rehabilitation training through the motor encoding vector can be achieved in the following way: First, the motor encoding vector can be normalized using a max-min normalization algorithm to eliminate the influence of individual differences and signal amplitude; then, the normalized motor encoding vector can be converted into specific joint motion trajectory parameters, including target position, motion velocity, and acceleration curve, using an existing kinematic mapping model (such as a musculoskeletal inverse kinematics model based on OpenSim), while simultaneously calculating a confidence index of the motor intention to evaluate the reliability of the decoding results; finally, the set containing the motion trajectory parameters and the confidence index is used as a description of the target patient's motor intention during rehabilitation training; wherein, the motor intention is a set of motor instructions describing the rehabilitation movement trajectory that the target patient expects to execute. This motor intention transforms abstract motor ideas into quantifiable motor parameters that the controller can execute, ensuring that rehabilitation training strictly follows the patient's active movement intention, which is the core of achieving active rehabilitation.

[0078] In specific implementation, determining the force contribution of each muscle group in the target patient based on the electromyography-torque mapping relationship and the coordinated movement state can be achieved in the following way: An electromyography-torque mapping model can be constructed using a long short-term memory network (LSTM). The memory units of the LTM network capture the temporal dependency between muscle activation and torque output. During the model training phase, a six-dimensional force sensor is used to measure the target patient's actual joint torque data as a supervision signal, and the network parameters are optimized using a backpropagation algorithm. Then, hierarchical feature selection and attention mechanisms can be used to filter out the muscle groups in the coordinated movement state that contribute to the force of each muscle group in the target patient's affected limb. The electromyographic features related to muscle groups are used to predict the contribution ratio of each muscle group to the total joint torque by inputting these features into a trained electromyographic-torque mapping model. A sliding window mechanism is employed to smooth the prediction results, ultimately outputting the force contribution of each muscle group. This yields the force contribution of each muscle group in the target patient. The force contribution is a parameter that quantifies the proportion of torque generated by the muscle groups in the affected limb during the overall movement of the target patient relative to the total target torque. This force contribution allows for precise quantification of the function of individual muscles in multi-muscle group synergy, providing data support for identifying dominant and accessory muscles and assessing the presence of abnormal compensation.

[0079] It should be noted that, in this application, the muscle strength contribution distribution is a collaborative pattern atlas reflecting the relationship between the activation sequence and intensity of various muscles during limb movement in the target patient during rehabilitation training. This muscle strength contribution distribution characterizes the coordination strategy of the neuromuscular system during movement, including the spatiotemporal characteristics of muscle activation and the preference for using collaborative patterns. It can visualize and quantify complex neuromuscular control strategies, providing a key evaluation tool for assessing the correctness of movement patterns and developing targeted muscle retraining programs. In specific implementation, the muscle strength contribution distribution of the target patient during rehabilitation training based on the force contribution of all muscle groups can be achieved in the following way: First, the force contribution of each muscle group can be organized into a muscle strength contribution matrix according to the time sequence. The rows of this muscle strength contribution matrix represent different muscle groups, the columns represent time points, and the element values ​​reflect the activation intensity of each muscle group at different time points. Then, a non-negative matrix factorization algorithm can be used to extract the collaborative patterns of the dominant movement execution and calculate the weight coefficients and activation sequences of each collaborative pattern, thereby generating a muscle strength collaborative activation atlas, which is then used as the muscle strength contribution distribution of the target patient during rehabilitation training.

[0080] In step 103, a dynamic risk map is determined for the target patient during rehabilitation training by considering the interaction between the target patient and the rehabilitation training robot in the rehabilitation training environment and the spatial domain occupied by obstacles.

[0081] In some embodiments, reference Figure 3 The figure is an exemplary flowchart illustrating the determination of a dynamic risk map according to some embodiments of this application. The determination of the dynamic risk map of a target patient during rehabilitation training in this application, based on the interaction state between the target patient and the rehabilitation training robot in the rehabilitation training environment and the spatial domain occupied by obstacles, can be achieved through the following steps:

[0082] In step 1031, the spatial position of obstacles in the rehabilitation training environment is monitored in real time to obtain the motion state of the obstacles;

[0083] In step 1032, the spatial domain occupied by the obstacle is determined based on the motion state;

[0084] In step 1033, the relative pose relationship between the target patient and the rehabilitation training robot in the rehabilitation training environment is captured, thereby determining the interaction state between the target patient and the rehabilitation training robot.

[0085] In step 1034, multi-source perception fusion is performed on the spatial domain and the interaction state to establish a dynamic risk map of the rehabilitation training environment.

[0086] In practical implementation, real-time monitoring of the spatial positions of obstacles in the rehabilitation training environment and obtaining the motion state of the obstacles can be achieved in the following way: First, a 77GHz millimeter-wave radar can be used as an environmental sensor to transmit frequency-modulated continuous waves in the rehabilitation training environment and receive reflected signals. At the same time, the distance, velocity, and angle information of the obstacles are calculated using fast Fourier transform, thereby obtaining the state data of all obstacles. Then, a density-based spatial clustering algorithm can be used to cluster the state data to distinguish different obstacles, and the motion trajectory of each obstacle can be predicted and updated using a Kalman filter algorithm. Finally, the data set of the three-dimensional position coordinates, motion velocity, acceleration, and motion direction of all obstacles is output as the motion state of the obstacles. The motion state is a dynamic information set describing the spatial position and motion vector of obstacles in the rehabilitation training environment. This motion state can realize full-dimensional perception and trajectory prediction of dynamic obstacles in the rehabilitation environment, providing a data basis for forward-looking safety decisions. Moreover, the millimeter-wave radar solution has advantages such as resistance to ambient light interference and strong penetration, and can reliably detect various dynamic obstacles such as medical staff and mobile equipment that may appear in the rehabilitation environment.

[0087] In specific implementation, determining the spatial domain occupied by obstacles based on the motion state can be achieved in the following way: the rehabilitation environment can be discretized into a three-dimensional grid using a spatiotemporal probability occupancy grid method. Each grid is assigned a probability value of being occupied by an obstacle. A linear Gaussian motion model is used to predict the possible position distribution of each obstacle within a future time window based on the motion state of the obstacles. A confidence interval (e.g., 95%) is then set, and grids with probability values ​​exceeding the set confidence interval are marked as the spatial domain occupied by obstacles, thus obtaining the spatial domain occupied by obstacles. Here, the spatial domain represents the extended safety boundary occupied by obstacles in the physical space of the rehabilitation training environment in the current and short future period. This spatial domain can extend the physical outline of the obstacle into a dynamic safety buffer containing spatiotemporal uncertainties, realizing the transition from passive collision avoidance to active obstacle avoidance. In addition, this dynamic occupancy domain not only includes the current volume of the obstacle but also the predicted extended area in its motion direction, providing forward-looking information for safety decisions.

[0088] In practical implementation, capturing the relative pose relationship between the target patient and the rehabilitation training robot in the rehabilitation training environment, and thus determining the interaction state between the target patient and the rehabilitation training robot, can be achieved in the following way: First, a binocular stereo vision camera can be used to collect the environmental depth information of the target patient during rehabilitation training. Then, based on an open-source pose detection algorithm, the pose timing of key points on the target patient's body (e.g., shoulder, elbow, wrist, etc.) can be detected using this environmental depth information and limb motion data. Next, the pose timing of the rehabilitation robot's end effector is calculated using the robot's forward kinematics model, and the coordinates of the target patient's joints and the robot's end effector are transformed to the same coordinate system. After establishing the coordinate system, the Euclidean distance and approach angle between the target patient and the corresponding end effector of the rehabilitation training robot are calculated using the pose and timing sequence of the key points of the target patient's body and the pose and timing sequence of the corresponding end effector. The vector set composed of the Euclidean distances and approach angles between all key points of the body and the corresponding end effectors is taken as the interaction state between the target patient and the rehabilitation training robot. The interaction state is a vector set that quantifies the relative pose and kinematic relationship between the target patient and the rehabilitation robot. This interaction state can characterize the real-time relative relationship between the human-machine system during rehabilitation training, providing a core basis for assessing the collaborative safety and potential conflict risks between the two in the shared workspace.

[0089] In specific implementation, the multi-source perception fusion of the spatial domain and the interaction state to establish a dynamic risk map of the rehabilitation training environment can be achieved in the following way: First, the Dempster-Schafer evidence theory can be used to fuse the spatial domain and the interaction state to obtain a dynamic risk map of the rehabilitation training environment. That is, inference rules based on fuzzy logic can be established, for example: if the obstacle is close and the approach speed is fast, the risk is high. Then, the rehabilitation training environment is divided into fine-grained units through gridding processing, and the obstacle distribution probability provided by the spatial domain and the human-computer relationship data provided by the interaction state are used as independent evidence sources to calculate the risk confidence of each fine-grained unit, thereby generating a situation map with different color depths representing risk levels; then... By performing integral calculations on the risk hotspot areas in the situation map, the overall environmental risk value and the regional environmental risk value are calculated, i.e., the sum of the products of the risk value and the area. The final output situation map containing the overall environmental risk value and the regional environmental risk value serves as a dynamic risk map of the rehabilitation training environment. The dynamic risk map is an overall safety distribution map that presents the real-time risk levels of different areas in the rehabilitation training environment. This dynamic risk map can intuitively display the safety status of each area in the rehabilitation training environment, especially the risk hotspots in the human-machine-environment interaction area. This simplifies the complex multi-source environmental information into an intuitive global risk heat map, enabling the system to quickly identify risk hotspots and formulate optimal collaborative safety strategies.

[0090] In step 104, variable domain fuzzy reasoning is performed on the movement intention, the muscle strength contribution distribution, and the dynamic risk map, and then rehabilitation adjustment instructions matching the rehabilitation status of the target patient are determined based on the reasoning results.

[0091] In some embodiments, performing variable-domain fuzzy reasoning on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, and then determining rehabilitation adjustment instructions matching the rehabilitation status of the target patient based on the reasoning results, can be achieved through the following steps:

[0092] Construct a multidimensional rehabilitation decision space with fuzzy reasoning between clarity of motor intent, muscle strength coordination efficiency, and environmental risk level;

[0093] Through the adaptive domain adjustment mechanism of the rehabilitation stage in the multidimensional rehabilitation decision space, combined with the real-time movement deviation of the target patient, the variable domain scaling factor of multimodal collaborative reasoning is determined according to the movement intention, the muscle strength contribution distribution and the dynamic risk map.

[0094] Based on the variable universe scaling factor, the universe of discourse of the fuzzy inference is corrected, and then the intensity correction coefficient and the auxiliary force adjustment coefficient of the rehabilitation training are obtained.

[0095] Based on the intensity correction coefficient and the auxiliary force adjustment coefficient, a rehabilitation adjustment instruction matching the rehabilitation status of the target patient is generated.

[0096] In practical implementation, the construction of a multidimensional rehabilitation decision space for fuzzy reasoning between motor intention clarity, muscle strength coordination efficiency, and environmental risk level can be achieved in the following way: First, three input variables can be established, including membership functions for motor intention clarity, muscle strength coordination efficiency, and environmental risk level. Specifically, motor intention clarity can be defined by the signal-to-noise ratio and stability of the cortical motor coding vector, using a Gaussian membership function; muscle strength coordination efficiency can be calculated by the coordination coefficient of the principal antagonist muscle and the synchronicity of force exertion, using a trapezoidal membership function; and environmental risk level can be determined by the risk field strength integral of the dynamic risk map, using a triangular membership function. Then, a hierarchical fuzzy rule base containing 27 rules is constructed. The first layer has 9 rules for safety assessment, and the second layer has 18 rules for training effect optimization. For example, if the clarity of intention is high, muscle strength coordination efficiency is moderate, and environmental risk is low, the training intensity is increased. This ultimately forms a multidimensional rehabilitation decision space that includes membership functions and a hierarchical fuzzy rule base. The multidimensional rehabilitation decision space is a comprehensive reasoning framework that integrates the target patient's movement intention, muscle function status, and environmental safety information. This multidimensional rehabilitation decision space can achieve a leap from single-dimensional decision-making to multi-factor collaborative decision-making, ensuring that the rehabilitation training program simultaneously meets the requirements of initiative, effectiveness, and safety. The fuzzy reasoning parameters in the multidimensional rehabilitation decision space can be calibrated based on clinical rehabilitation assessment standards, such as muscle strength level and joint range of motion.

[0097] In specific implementation, the adaptive domain adjustment mechanism of the rehabilitation stage in the multidimensional rehabilitation decision space, combined with the real-time movement deviation of the target patient, determines the variable domain scaling factor of multimodal collaborative reasoning based on the movement intention, the muscle strength contribution distribution, and the dynamic risk map. This can be achieved in the following way: First, establish a domain benchmark based on the rehabilitation stage, dividing the rehabilitation process into three stages: early, middle, and late, with different initial domain ranges set for each stage; then, calculate in real time the positional deviation, velocity deviation, and acceleration deviation between the target patient's executed actions and the standard rehabilitation trajectory, and combine the positional deviation, velocity deviation, and acceleration deviation into a vector to obtain a comprehensive movement deviation index; finally, construct a multimodal parameter fusion model to integrate the activation stability of the principal muscle groups in the muscle strength contribution distribution, the environmental risk value in the dynamic risk map, and the movement intention. The confidence index is used as three independent input dimensions. Normalized parameter values ​​for each dimension are calculated, and a weighted geometric mean algorithm is used to fuse the parameter values ​​of the three dimensions to obtain a fused parameter value. The weight coefficients of each dimension are dynamically adjusted according to the rehabilitation stage. In the early stage, the dynamic risk map is given a higher weight to ensure safety, and in the later stage, the movement intention is given a higher weight to promote active rehabilitation. Finally, the fused parameter value is multiplied by the movement deviation index and then multiplied by the stage adjustment coefficient to obtain the variable domain scaling factor of the multimodal collaborative inference. The variable domain scaling factor is an adaptive parameter that dynamically adjusts the perception accuracy of the fuzzy inference system according to the real-time ability of the target patient and changes in the environment. This variable domain scaling factor can overcome the defects of fixed accuracy in traditional fuzzy control, enabling the system to not only keenly capture subtle functional changes, but also robustly respond to sudden risks.

[0098] In specific implementation, the intensity correction coefficient and auxiliary force adjustment coefficient for rehabilitation training can be obtained by modifying the domain range of fuzzy inference based on the variable domain scaling factor, as follows: First, the variable domain scaling factor can be applied to the domain boundary of the input and output variables in the multidimensional rehabilitation decision space to achieve real-time adjustment of the domain range; then, fuzzy inference is performed within the new domain range, using the Mamdani inference method to obtain the output fuzzy set through max-min composition operation; finally, the centroid method is used for defuzzification calculation to obtain the accurate intensity correction coefficient and auxiliary force adjustment coefficient; wherein, the intensity correction coefficient... The intensity correction coefficient is a quantitative proportional factor that dynamically adjusts the difficulty of the training task based on the real-time performance of the target patient during rehabilitation training. The range of this intensity correction coefficient is set to [0.5, 1.5], which can achieve adaptive and precise control of the intensity of rehabilitation training and ensure that the training is always at the best balance between challenge and completion. The assistance force adjustment coefficient is a standardized parameter that quantifies the amount of external assistance provided by the rehabilitation robot. The range of this assistance force adjustment coefficient is set to [0.3, 1.0], which is used to control the amount of assistance provided by the rehabilitation robot. It can realize the transformation from an all-or-nothing assistance mode to a fine assistance mode that is allocated on demand, effectively promoting the remodeling of neural function.

[0099] It should be noted that in this application, the rehabilitation adjustment instruction is a set of executable instructions that integrates all control parameters during rehabilitation training to drive the rehabilitation robot to perform personalized training. This rehabilitation adjustment instruction can transform complex clinical decisions into standardized operations that the robot can directly execute, bridging the last mile from assessment to intervention. Specifically, the generation of rehabilitation adjustment instructions matching the target patient's rehabilitation status based on the intensity correction coefficient and the auxiliary force adjustment coefficient can be achieved in the following way: First, new training parameters can be calculated based on the intensity correction coefficient, namely: trajectory amplitude = baseline amplitude × intensity coefficient, movement speed = baseline speed × intensity coefficient to the power of 0.5, training duration = baseline duration × intensity coefficient; then, the target impedance parameters of each joint of the rehabilitation robot can be calculated based on the auxiliary force adjustment coefficient, namely: stiffness = maximum stiffness × adjustment coefficient, damping = maximum damping × adjustment coefficient to the power of 1.5; at the same time, safety constraints are generated by combining a dynamic risk map, including: maximum range of motion limit, speed limit and obstacle avoidance priority, and finally outputting a set of rehabilitation adjustment instructions containing training parameters, target impedance parameters and safety constraints.

[0100] In step 105, the rehabilitation progress of the target patient is continuously tracked and the rehabilitation training robot is controlled to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

[0101] In some embodiments, continuously tracking the rehabilitation progress of the target patient and controlling the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions can be achieved by the following steps:

[0102] Based on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, a dynamic rehabilitation efficacy index is constructed to continuously track the rehabilitation progress of the target patient.

[0103] Based on the rehabilitation adjustment instructions and the dynamic rehabilitation efficacy indicators, the rehabilitation training robot is controlled to provide impedance-adaptive dynamic assistance compensation to the target patient.

[0104] In specific implementation, the construction of a dynamic rehabilitation efficacy index based on the motor intention, the muscle strength contribution distribution, and the dynamic risk map to continuously track the rehabilitation progress of the target patient can be achieved in the following way: First, the average value of all confidence indicators in the motor intention can be used as the functional independence score during the rehabilitation training process of the target patient. Then, the maximum muscle strength value in the muscle strength contribution distribution can be converted into a standard muscle strength level through the existing muscle strength level mapping model. Finally, the overall environmental risk value in the dynamic risk map can be used as the environmental adaptation index during the rehabilitation training process of the target patient, thereby constructing a dynamic rehabilitation efficacy index that includes the functional independence score, standard muscle strength level, and environmental adaptation index. Then, the short-term trend and long-term rate of change of each efficacy index can be calculated through a sliding window mechanism to form a complete dynamic tracking curve of rehabilitation progress to continuously track the rehabilitation progress of the target patient. The dynamic rehabilitation efficacy index is a comprehensive performance indicator that quantifies the quality and progress of rehabilitation training of the target patient. It includes the functional independence score, standard muscle strength level, and environmental adaptation index. This dynamic rehabilitation efficacy index can transform the complex rehabilitation process into traceable and optimizable objective data, realizing the transformation from subjective experience judgment to objective data-driven rehabilitation management.

[0105] In specific implementation, the dynamic assistance compensation for impedance adaptation of the rehabilitation training robot to the target patient, based on the rehabilitation adjustment instructions and the dynamic rehabilitation efficacy index, can be achieved in the following way: First, establish an impedance adaptive adjustment mechanism based on admittance control. By analyzing the target impedance parameters in the rehabilitation adjustment instructions, including stiffness coefficient, damping coefficient, and inertia parameter, and combining them with the muscle strength level and functional independence score in the dynamic rehabilitation efficacy index, a parameter mapping algorithm is used to calculate the target impedance value suitable for the current rehabilitation stage in real time. Then, construct an impedance control architecture based on the position inner loop. The human-machine interaction force information is collected in real time through a six-dimensional force sensor, and the expected motion trajectory correction of the robot end is calculated based on the admittance control law. This control law calculates the ratio of the contact force to the target impedance parameter. The values ​​are converted into trajectory adjustment quantities to achieve compliant control. Based on this, a variable universe of discourse fuzzy proportional-integral-derivative controller is introduced. The proportional coefficient, integral coefficient, and derivative coefficient of the controller are adjusted in real time according to the changing trend of dynamic rehabilitation efficacy indicators to ensure stable control performance at different rehabilitation stages. At the same time, a safety monitoring mechanism is established. When the interaction force exceeds the safety threshold or the motion trajectory is abnormal, it automatically switches to admittance control mode and reduces the stiffness coefficient in the impedance parameter to ensure the safety of the training process. Finally, the desired trajectory command after impedance adaptive adjustment is converted into motion control signals for each joint of the robot. Precise torque output is achieved through servo drives. This impedance control strategy based on multi-source information fusion is used as the core technical means to achieve personalized and adaptive rehabilitation training.

[0106] In another aspect, in one embodiment, this application provides a patient rehabilitation tracking system, which includes a rehabilitation training control unit, with reference to... Figure 4 The figure is a schematic diagram of the structure of a rehabilitation training control unit 400 according to some embodiments of this application. The rehabilitation training control unit 400 includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below:

[0107] The acquisition module 401 in this application is mainly used to synchronously acquire limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the rehabilitation training of the target patient.

[0108] Processing module 402 in this application is mainly used to determine the coordinated movement state of the neuromusculoskeletal system in the rehabilitation training of the target patient based on the multimodal physiological data stream, and then perform intention-muscle force coordination identification based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient based on the decoding of neurophysiological signals during rehabilitation training.

[0109] It should be noted that the processing module 402 in this application is also used to determine the dynamic risk map of the target patient when undergoing rehabilitation training by the interaction state between the target patient and the rehabilitation training robot and the spatial domain occupied by obstacles in the rehabilitation training environment.

[0110] Additionally, it should be noted that the processing module 402 in this application is also used to perform variable domain fuzzy reasoning on the motor intention, the muscle strength contribution distribution and the dynamic risk map, and then determine the rehabilitation adjustment instructions that match the rehabilitation status of the target patient based on the reasoning results;

[0111] The execution module 403 in this application is mainly used to continuously track the rehabilitation progress of the target patient and control the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

[0112] The modules in the aforementioned patient rehabilitation tracking system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0113] In another embodiment, this application provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 5 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data on a rehabilitation training and control method based on active perception and environmental adaptation. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program can implement a rehabilitation training and control method based on active perception and environmental adaptation.

[0114] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0115] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps described in the embodiment of the rehabilitation training regulation method based on active perception and environmental adaptation.

[0116] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps described in the embodiment of the rehabilitation training regulation method based on active perception and environmental adaptation.

[0117] In one embodiment, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps described in the embodiment of the rehabilitation training modulation method based on active perception and environmental adaptation.

[0118] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0119] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A rehabilitation training regulation method based on active perception and environmental adaptation, used in a patient rehabilitation tracking system for adaptive rehabilitation training regulation of target patients, characterized in that, Includes the following steps: Simultaneously collect limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, and then obtain a multimodal physiological data stream for tracking the rehabilitation training of the target patient; Based on the multimodal physiological data stream, the coordinated movement state of the neuromusculoskeletal system during the rehabilitation training of the target patient is determined, and then the intention-muscle force coordination identification is performed based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient based on the decoding of neurophysiological signals during rehabilitation training. A dynamic risk map of the target patient during rehabilitation training is determined by the interaction between the target patient and the rehabilitation training robot in the rehabilitation training environment and the spatial domain occupied by obstacles. The exercise intention, the muscle strength contribution distribution and the dynamic risk map are subjected to variable domain fuzzy reasoning, and then the rehabilitation adjustment instructions that match the rehabilitation status of the target patient are determined based on the reasoning results. Continuously track the rehabilitation progress of the target patient and control the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions; Specifically, the process of performing variable-domain fuzzy reasoning on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, and then determining rehabilitation adjustment instructions to match the rehabilitation status of the target patient based on the reasoning results, includes: Construct a multidimensional rehabilitation decision space with fuzzy reasoning between clarity of movement intent, muscle strength coordination efficiency, and environmental risk level; Through the adaptive domain adjustment mechanism of the rehabilitation stage in the multidimensional rehabilitation decision space, combined with the real-time movement deviation of the target patient, the variable domain scaling factor of multimodal collaborative reasoning is determined according to the movement intention, the muscle strength contribution distribution and the dynamic risk map. Based on the variable universe scaling factor, the universe of discourse of the fuzzy inference is corrected, and then the intensity correction coefficient and the auxiliary force adjustment coefficient of the rehabilitation training are obtained. Based on the intensity correction coefficient and the auxiliary force adjustment coefficient, a rehabilitation adjustment instruction matching the rehabilitation status of the target patient is generated.

2. The rehabilitation training regulation method based on active perception and environmental adaptation as described in claim 1, characterized in that, Simultaneously collecting limb movement data and electromyographic signals of the target patient during rehabilitation training tasks, and thus obtaining a multimodal physiological data stream for tracking the target patient's rehabilitation training, specifically includes: The inertial measurement unit collects limb movement data of the target patient when performing rehabilitation training tasks; Electromyographic signals of target patients during rehabilitation training tasks were acquired using a surface electromyography sensor array. The limb movement data and the electromyographic signals are time-stamped to form a multimodal physiological data stream for tracking the rehabilitation training of the target patient.

3. The rehabilitation training regulation method based on active perception and environmental adaptation as described in claim 1, characterized in that, Determining the coordinated motor state of the neuromuscular system during rehabilitation training of the target patient based on the aforementioned multimodal physiological data stream specifically includes: Logarithmic time-domain features of electromyographic physiological signals in the multimodal physiological data stream are extracted, and then an electromyographic feature matrix is ​​constructed. Extract joint angle information of the target patient during limb movement from the multimodal physiological data stream; By spatiotemporally coupling the electromyographic feature matrix with the joint angle information, the coordinated movement state of the neuromuscular system during the rehabilitation training of the target patient can be obtained.

4. The rehabilitation training regulation method based on active perception and environmental adaptation as described in claim 1, characterized in that, Based on the aforementioned coordinated movement state, intention-muscle coordination recognition is performed to obtain the distribution of movement intention and muscle strength contribution based on neurophysiological signal decoding during rehabilitation training of the target patient. Specifically, this includes: By combining the target patient's electroencephalogram (EEG) signals, the coordinated motor state is decoded hierarchically to obtain the motor coding vector of the target patient's motor cortex; The motion encoding vector is used to determine the target patient's motion intention based on the decoding of neurophysiological signals during rehabilitation training; Based on the electromyography-torque mapping relationship of the target patient, the force contribution of each muscle group in the target patient is determined according to the synergistic motion state; The distribution of muscle force contribution during rehabilitation training for the target patient is generated based on the force contribution of all muscle groups.

5. The rehabilitation training regulation method based on active perception and environmental adaptation as described in claim 1, characterized in that, The dynamic risk map for the target patient during rehabilitation training is determined by analyzing the interaction between the target patient and the rehabilitation training robot within the rehabilitation training environment and the spatial domain occupied by obstacles. Specifically, this includes: Real-time monitoring of the spatial location of obstacles in the rehabilitation training environment to obtain the motion status of the obstacles; The spatial domain occupied by the obstacle is determined based on the motion state; Capture the relative pose relationship between the target patient and the rehabilitation training robot in the rehabilitation training environment, and then determine the interaction state between the target patient and the rehabilitation training robot; Multi-source perception fusion is performed on the spatial domain and the interaction state to establish a dynamic risk map of the rehabilitation training environment.

6. The rehabilitation training regulation method based on active perception and environmental adaptation as described in claim 1, characterized in that, Continuously tracking the rehabilitation progress of the target patient and controlling the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions specifically includes: Based on the motor intention, the muscle strength contribution distribution, and the dynamic risk map, a dynamic rehabilitation efficacy index is constructed to continuously track the rehabilitation progress of the target patient. Based on the rehabilitation adjustment instructions and the dynamic rehabilitation efficacy indicators, the rehabilitation training robot is controlled to provide impedance-adaptive dynamic assistance compensation to the target patient.

7. A patient rehabilitation tracking system, comprising a rehabilitation training control unit, wherein the rehabilitation training is controlled using the method described in any one of claims 1 to 6, characterized in that, The rehabilitation training control unit includes: The acquisition module is used to synchronously acquire limb movement data and electromyographic signals of the target patient when performing rehabilitation training tasks, thereby obtaining a multimodal physiological data stream for tracking the rehabilitation training of the target patient; The processing module is used to determine the coordinated movement state of the neuromusculoskeletal system during the rehabilitation training of the target patient based on the multimodal physiological data stream, and then perform intention-muscle force coordination identification based on the coordinated movement state to obtain the movement intention and muscle force contribution distribution of the target patient based on the decoding of neurophysiological signals during rehabilitation training. The processing module is used to determine a dynamic risk map of the target patient during rehabilitation training by using the interaction state between the target patient and the rehabilitation training robot and the spatial domain occupied by obstacles in the rehabilitation training environment. The processing module is used to perform variable domain fuzzy reasoning on the movement intention, the muscle strength contribution distribution and the dynamic risk map, and then determine rehabilitation adjustment instructions that match the rehabilitation status of the target patient based on the reasoning results. The execution module is used to continuously track the rehabilitation progress of the target patient and control the rehabilitation training robot to perform adaptive assistance compensation according to the rehabilitation adjustment instructions.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the rehabilitation training regulation method based on active perception and environmental adaptation as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the rehabilitation training regulation method based on active perception and environmental adaptation as described in any one of claims 1 to 6.

Images