Muscle injury rehabilitation treatment method and system based on multi-modal data fusion

Abstract

The invention relates to a muscle injury rehabilitation treatment method and system based on multi-modal data fusion. The method comprises the following steps: S1, acquiring data including myoelectricity, electroencephalogram and posture data; s2, performing fusion evaluation based on the multi-modal data to obtain a multi-modal evaluation score; s3, generating and executing an individual rehabilitation treatment scheme based on the multi-modal evaluation score; and S4, when it is detected that the rehabilitation state reaches a preset threshold value, the steps S1 and S2 are triggered and executed again, and the individual rehabilitation treatment scheme is updated based on the newly generated multi-modal evaluation score. According to the method, accurate assessment of the muscle contraction state and the injury degree is achieved through multi-dimensional data fusion, a dynamic treatment scheme is generated in combination with a personalized adaptation mechanism, and the rehabilitation process is tracked and optimized through full-cycle data. The problems that an existing rehabilitation technology is single in evaluation dimension, insufficient in personalized adaptation and lack of dynamic adjustment of a treatment scheme are solved, and the method is suitable for rehabilitation scenes of damage of various tissues such as bones, ligaments, cartilages and skeletal muscles.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of rehabilitation medicine, biomedical engineering, and computer technology, and specifically to a method and system for muscle injury rehabilitation based on multimodal data fusion. Background Technology

[0002] With the rapid development of medical-engineering interdisciplinary technologies, rehabilitation therapy is undergoing a profound transformation from a traditional, experience-based model towards a data-driven, precise, and personalized approach. Brain-computer interfaces, electromyography (EMG), and computer vision technologies provide new tools for rehabilitation assessment and intervention. However, existing technologies still have significant limitations in clinical application, restricting their full effectiveness.

[0003] First, current technologies largely focus on single-function rehabilitation scenarios or the application of single-modal data. For example, surface electromyography (EMG) is used alone to assess muscle activation, or motion capture systems are used alone to assess joint range of motion. This single-dimensional assessment is insufficient to comprehensively reflect the patient's true rehabilitation status. Muscle injury rehabilitation involves not only the restoration of the muscle's physiological functions (such as strength and endurance), but also the reconstruction of motor control patterns (such as coordination and stability) and the regulation of neuropsychological factors (such as pain tolerance and attention training). No single modality can cover all these aspects, leading to one-sided assessment results, and treatment plans developed based on these results often fail to address all aspects of the problem.

[0004] Secondly, existing technologies are insufficient for the collaborative application of multimodal data, particularly lacking mechanisms for effectively fusing asynchronous, non-uniformly periodic multi-source data. Electromyography (EMG) signals have high sampling frequencies (often several kilohertz) and change rapidly; electroencephalography (EEG) signals reflect relatively slower changes in neuropsychological states; and vision-based posture data is limited by camera frame rates and algorithm processing latency. These data are inherently asynchronous and periodically inconsistent. Simple time-series splicing or independent analysis followed by merging introduces temporal mismatches, leading to distorted fusion evaluation results that fail to accurately reflect the coordinated state of the "muscle-nerve-joint" system at the same moment. Achieving high-precision temporal alignment and reliability-weighted fusion of multimodal data is a key technical challenge that existing technologies have not adequately addressed.

[0005] Furthermore, existing rehabilitation systems generally lack personalization and dynamic adaptability. Many systems train based on fixed, standardized thresholds or protocols, failing to fully consider individual patient differences, such as injury type and severity, postural characteristics, basic muscle strength, and pain sensitivity. At the same time, once a treatment plan is generated, it often lacks the ability to dynamically adjust based on real-time performance, failing to provide precise guidance and parameter optimization "according to the situation" during training, resulting in low training efficiency and potentially causing secondary injuries due to improper training.

[0006] In addition, the scope of applicable diseases for existing technologies is relatively narrow, mainly focusing on neurogenic rehabilitation such as hemiplegia after stroke and spinal cord injury or rehabilitation after routine limb fracture surgery. There is less support for in-depth and precise rehabilitation in areas such as complex skeletal muscle injury, ligament and cartilage injury, and muscle dysfunction associated with neurodegenerative diseases, which makes it difficult to meet the increasingly diversified clinical rehabilitation needs.

[0007] In summary, there is an urgent need for a muscle injury rehabilitation technology solution that can integrate multi-dimensional information, achieve precise asynchronous fusion assessment, and provide personalized dynamic treatment throughout the entire treatment cycle. Summary of the Invention

[0008] The present invention aims to overcome the above-mentioned defects of the prior art and provide a muscle injury rehabilitation treatment method and system based on multimodal data fusion. Its core lies in constructing a closed-loop intelligent rehabilitation system of "collection-assessment-treatment-tracking". Through innovative multimodal asynchronous data fusion assessment and personalized dynamic adaptation mechanism, the scientific, precise and intelligent rehabilitation of muscle injuries can be achieved.

[0009] The first aspect of this invention discloses a muscle injury rehabilitation method based on multimodal data fusion, comprising the following steps: S1, collecting electromyographic signals, electroencephalogram (EEG) signals, and posture data of the target subject and preprocessing them to obtain electromyographic feature data, EEG index data, and posture parameter data, forming multimodal data; S2, performing fusion evaluation based on the multimodal data to obtain a multimodal evaluation score; S3, generating and executing an individual rehabilitation treatment plan based on the multimodal evaluation score and the target subject's personalized adaptation parameters; S4, recording the target subject's rehabilitation progress data and judging the rehabilitation status based on it. When the rehabilitation status is detected to reach a preset threshold, steps S1 and S2 are triggered and re-executed, and the individual rehabilitation treatment plan is updated based on the newly generated multimodal evaluation score.

[0010] According to the method disclosed in the first aspect of the present invention, step S2 includes: S21, extracting the timestamps of each of the multimodal data, and mapping the non-uniformly sampled multimodal data to a preset reference time axis through an interpolation algorithm, so that each reference time point can correspond to a set of multimodal data; S22, assigning a sampling period adaptation coefficient to each of the multimodal data mapped to the reference time axis; S23, constructing a three-dimensional modality assessment model including electromyography modality, electroencephalography modality and posture modality, and assigning modality weights to each modality, and assigning modality weights to the corresponding modalities. The feature indicators are assigned sub-weights respectively, and the sum of the sub-weights of all feature indicators under each mode is equal to the mode weight of the corresponding mode; S24, based on the corresponding data of the feature indicators under each mode and the sub-weights of the corresponding feature indicators, the single-mode evaluation value of the corresponding mode is calculated respectively, and each obtained single-mode evaluation value is multiplied with the sampling period adaptation coefficient assigned to the corresponding mode in step S22 to perform credibility modulation; S25, the modal evaluation values ​​after credibility modulation are weighted and summed according to the corresponding assigned modal weights to obtain the multimodal evaluation score.

[0011] Step S21 is the time-series alignment step. Addressing the asynchronous sampling periods of EMG, EEG, and posture data, a high-precision timestamp is first applied to all data. Then, using a preset, high-update-rate baseline time axis (e.g., a 10ms interval) as a reference, algorithms such as linear interpolation are used to "align" the data across modalities in the time dimension. This ensures that at each baseline time point, theoretically synchronized EMG, EEG, and posture data pairs are obtained. This solves the primary obstacle to multimodal data fusion—time consistency.

[0012] Step S22 is the dynamic adaptation step, and S24 is the credibility modulation step. Those skilled in the art will understand that not all aligned data has equal credibility. To address this issue, this invention innovatively introduces a "sampling period adaptation coefficient K," which is dynamically assigned based on the actual sampling characteristics of each data source (e.g., K≈1.0 for electromyography, K=0.6-0.8 for posture). Data sources with high sampling frequencies and stable periods (e.g., electromyography) have stronger timeliness and higher credibility. That is, when calculating the evaluation value of each modality, it is multiplied by the corresponding K value to achieve "credibility modulation." High-frequency reliable data is enhanced, while the weight of low-frequency or highly fluctuating data is relatively suppressed.

[0013] Steps S23 and S25 involve the construction and weighted fusion of three-dimensional modalities. Specifically, a three-dimensional assessment model is constructed, comprising "electromyographic modality, electroencephalography modality, and postural modality." The assessment value for each modality is calculated by weighting multiple feature indicators under its sub-modality. Then, the modal assessment values, after reliability modulation, are weighted and summed according to pre-assigned, adjustable modal weights to obtain a multimodal assessment score ranging from 0 to 100. This score comprehensively reflects the patient's immediate state across the three levels of "muscle function, motor control, and neuropsychological aspects."

[0014] Based on this application, the multimodal fusion assessment achieves a scientific and robust fusion of multi-source asynchronous data into a single comprehensive score. Specifically, the dual mechanism of "temporal alignment-dynamic adaptation" not only solves the synchronization problem but also adaptively handles changes in data quality at different training stages (such as decreased EMG signal quality during fatigue) through the K-coefficient, ensuring stable and reliable assessment results across various scenarios. Furthermore, the dynamic adjustment of the weights in the three-dimensional modal model allows for adjustments. For example, in the early stages of rehabilitation, the weights of "postural modality" and "EEG modality" (pain) can be increased to emphasize safety and pain-free range of activities; in the later stages of rehabilitation, the weight of "EMG modality" can be increased to emphasize strength and explosive power training. This dynamic weighting strategy enables the assessment model to flexibly adapt to different rehabilitation stage goals, resulting in a more targeted assessment that is impossible with fixed-weight models.

[0015] According to the method disclosed in the first aspect of the present invention, the characteristic indicators of the electromyography modality include muscle contraction strength achievement rate, electromyography activation degree score and muscle contraction stability score; the characteristic indicators of the electroencephalography modality include attention score and pain association index; the characteristic indicators of the posture modality include joint range of motion achievement rate, movement stability score and movement trajectory deviation value.

[0016] According to the method disclosed in the first aspect of the present invention, step S3 includes: S31, matching a target rehabilitation movement from a pre-stored standard movement library containing multiple standard rehabilitation training movements and their corresponding biomechanical parameters based on the multimodal assessment score; S32, constructing at least one set of initial training parameters based on the standard movement parameters of the target rehabilitation movement and combined with the individual parameters of the target object to form an initial individual rehabilitation treatment plan; S33, during the execution of the treatment plan, collecting electromyographic signals, electroencephalographic signals, and posture data of the target object in real time, comparing and verifying the real-time data with the standard movement parameters and the expected performance values ​​set based on the initial training parameters; and dynamically adjusting the initial training parameters according to the comparison results to optimize the individual rehabilitation treatment plan.

[0017] In steps S31 and S32, suitable movements from the standard movement library are matched based on the multimodal assessment score (reflecting current ability), and combined with the aforementioned personalized adaptation parameters, specific initial training parameters (intensity, number of sets, angle, speed, etc.) are derived to form a personalized initial treatment plan. In step S33, multimodal data is continuously collected during patient training and compared with preset standard movement parameters (ideal mode) and expected performance values ​​based on the initial parameters (personalized goals). Once a deviation is detected (such as insufficient angle or compensatory muscle overactivation), the real-time feedback module immediately provides precise guidance through multiple channels (visual and auditory). At the same time, the system intelligently and dynamically fine-tunes the training parameters (such as reducing difficulty or extending rest periods) to ensure that training is always within a safe and effective "optimal challenge zone."

[0018] By utilizing the specific technical solution in step S3 of this application, a leap from "static prescription" to "dynamic navigation" in rehabilitation therapy has been achieved. Specifically, the combination of the "standard movement library" and "personalized adaptation parameters" produces a "chemical effect." It not only considers "what movements to perform" (standards) but also precisely specifies "to what extent" (individualized thresholds) and can "fine-tune in real time" (dynamic optimization) during execution. This three-tiered personalized guarantee greatly improves the accuracy and safety of training. The real-time feedback and dynamic parameter adjustment mechanism is like equipping the patient with an intelligent rehabilitation therapist, capable of instantly correcting incorrect movement patterns, effectively preventing the solidification of incorrect movements, and accelerating the relearning process of correct movement patterns—something that traditional follow-up videos or periodic review and adjustment cannot match.

[0019] According to the method disclosed in the first aspect of the present invention, the step of dynamically adjusting the initial training parameters based on the comparison results in step S33 includes: when real-time comparison verification finds that at least one of muscle contraction strength, joint range of motion, or movement stability has not reached the expected performance value, providing text, voice, or visual guidance through real-time feedback, and adjusting at least one of the initial training parameters of training intensity, movement difficulty, or training duration.

[0020] The system automatically records multi-dimensional data from each training session and generates a visualized rehabilitation progress curve using trend fitting algorithms (such as multinomial fitting). It sets intelligent improvement thresholds (e.g., a score improvement of ≥5 points over three consecutive training sessions). When the system detects significant improvement in rehabilitation status, it automatically triggers a comprehensive reassessment (S1-S2) to obtain the patient's latest functional baseline and update the treatment plan for the next stage. This achieves a closed-loop, data-driven rehabilitation management system throughout the entire cycle. An unexpected technological benefit is that this "threshold-triggered reassessment" mechanism avoids the problems of underassessment or overassessment that can occur with regular assessments. It only initiates reassessment and plan upgrades when the patient's function shows substantial progress, making the rehabilitation process more scientific and efficient, optimizing resource allocation, and truly achieving a "step-by-step" leap in the treatment plan, perfectly matching the patient's functional recovery curve.

[0021] According to the method disclosed in the first aspect of the present invention, the preprocessing includes hierarchical denoising and feature extraction of the electromyographic signal, assessment of attention concentration and pain correlation index of the electroencephalogram signal, and identification of joint key points and extraction of kinematic parameters of the posture data.

[0022] According to the method disclosed in the first aspect of the present invention, the steps of performing hierarchical denoising and feature extraction on the electromyography (EMG) signal in the preprocessing include: eliminating power frequency interference based on an adaptive bandwidth notch filter; calibrating the signal baseline based on an algorithm based on the signal time gradient to eliminate low-frequency drift; separating the EMG signal from the composite electromagnetic interference based on an independent component analysis algorithm combined with dynamic sparsity constraints, and reconstructing the signal; performing feature verification on the EMG signals before and after denoising to ensure the existence of key EMG features; and extracting the EMG amplitude, integral EMG, and root mean square value within a dynamic time window based on the denoised EMG signal.

[0023] Electromyography (EMG) signals are susceptible to interference from power frequency, baseline drift, and combined noise from electrocardiograms (ECGs). Layered denoising employs targeted algorithms to systematically remove interference: notch filters precisely filter out specific power frequency signals; baseline calibration algorithms eliminate slow drift; and independent component analysis (ICA) with dynamic sparsity constraints effectively separates statistically independent source signals (such as EMG, ECG, and motion artifacts). The sparsity characteristics of EMG signals (impulsive action potentials) enhance the separation effect. Finally, feature verification ensures that effective physiological information is not lost. Feature extraction quantifies the activation level (amplitude, iEMG) and quality (RMS) of muscles from the denoised signal.

[0024] Electroencephalogram (EEG) signals are used to assess neuropsychological states. By analyzing the energy or ratios of specific frequency bands, continuous brain waves are converted into quantifiable attention scores and pain-related indices.

[0025] Regarding the processing of posture data, computer vision technology (such as optimized models based on open-source frameworks like MediaPipe Pose) can be used to identify the coordinates of key human points in the video stream in real time, and then calculate kinematic parameters such as joint angles, motion trajectories, and stability, transforming visual information into a digital description of motion function.

[0026] Through preprocessing, raw, noisy, and heterogeneous biological signals and visual information are transformed into a clean, computable, and standardized feature data stream, providing high-quality input for subsequent fusion. An unexpected technical benefit is that, by combining ICA denoising with dynamic sparsity constraints, it can remarkably preserve weak pathological electromyographic features even under strong ECG interference, which is crucial for the sensitive assessment of early or mild muscle injuries. Simultaneously, the optimized pose recognition algorithm maintains high accuracy even under partial occlusion (such as braces or clothing), improving the system's clinical usability and robustness.

[0027] According to the method disclosed in the first aspect of the present invention, the personalized adaptation parameters for the target object are obtained through the following steps: collecting the target object's postural parameters and injury characteristics information, comparing them with evidence-based rehabilitation data from a population database, dynamically calibrating the training target parameters using an individual-population data comparison algorithm, and generating a training threshold range for the target object as the personalized adaptation parameters. That is, collecting the patient's postural characteristics (height, limb length) and injury characteristics, comparing them with a massive evidence-based database from a population database, and calculating a suitable personalized training threshold range (such as safe range of joint range of motion, target muscle strength percentage) for the patient using an individual-population data comparison algorithm. This allows the solution to move from a "standard" to an "individualized" approach.

[0028] According to the method disclosed in the first aspect of the present invention, before step S1, a step S0 is included to locate and determine the type of muscle injury in the target object to identify the target muscle tissue; the electromyography (EMG) signal is collected from the target muscle tissue. By identifying the target muscle tissue, subsequent EMG acquisition becomes more targeted. Specifically, by precisely locking onto the damaged muscle or muscle group, this ensures a high degree of consistency between the EMG electrode placement, the posture acquisition focus area, and the injury area, guaranteeing the relevance of the acquired data to the rehabilitation goal from the source and avoiding interference from irrelevant signals. Moreover, through precise positioning, the system can intelligently recommend different electrode acquisition modes (such as needle-type or surface electrodes) for deep and superficial muscles and automatically set initial sampling parameters, realizing the initial personalization of the acquisition plan and laying a solid foundation for subsequent accurate assessment.

[0029] The second aspect of this application also discloses a system for implementing the method disclosed in the first aspect of the present invention, comprising: a damage diagnosis module configured to locate and determine the type of muscle damage in a target object to identify the target muscle tissue; a data acquisition and processing module configured to include: an electromyography (EMG) detection module for acquiring EMG signals from the target muscle tissue; an electroencephalogram (EEG) acquisition module for acquiring EEG signals from the target object; a posture acquisition module for acquiring the human posture of the target object; and a preprocessing module for preprocessing the EMG signals, EEG signals, and posture data to obtain EMG feature data, EEG index data, and posture parameter data, thereby forming multimodal data. This system also includes: a fusion assessment module, which is constructed to perform fusion assessment based on multimodal data to obtain a multimodal assessment score; a plan generation module, which is constructed to generate and execute an individual rehabilitation treatment plan based on the multimodal assessment score and the personalized adaptation parameters of the target object; and a progress tracking and update management module, which records the rehabilitation progress data of the target object and judges the rehabilitation status based on it. When the rehabilitation status is detected to reach a preset threshold, the data acquisition and processing module and the fusion assessment module are triggered to work again to generate a new multimodal assessment score and update the individual rehabilitation treatment plan based on the newly generated multimodal assessment score.

[0030] This system is the hardware and software embodiment of the above-described method steps. Each module corresponds one-to-one with each step and sub-step of the method, working together to realize the described method flow. The modular design of the system facilitates integration, deployment, and maintenance.

[0031] In summary, this invention creatively addresses the core pain points of muscle injury rehabilitation—namely, the difficulty in multimodal data fusion, insufficient personalization, and the lack of dynamic adjustment—through a series of interconnected and progressively advanced technical solutions. The various technical features support each other, creating a synergistic effect and ultimately achieving significant technological advancements in more accurate assessment, more personalized treatment, more efficient rehabilitation, and a safer experience. This invention possesses significant clinical application value and market potential.

[0032] Beneficial Effects: In the muscle injury rehabilitation treatment method and system based on multimodal data fusion of this invention, precise assessment of muscle contraction state and injury degree is achieved through multi-dimensional data fusion. Combined with a personalized adaptation mechanism, dynamic treatment plans are generated, and the rehabilitation process is optimized through full-cycle data tracking. This solves the problems of existing rehabilitation technologies, such as single assessment dimensions, insufficient personalized adaptation, and lack of dynamic adjustment of treatment plans. It improves the scientific rigor, precision, and safety of rehabilitation training, and is applicable to rehabilitation scenarios involving various tissue injuries, including bones, ligaments, cartilage, and skeletal muscles. It is compatible with Google's open-source technology framework and can be widely applied in hospital, community, and home rehabilitation settings.

[0033] The following describes in detail the muscle injury rehabilitation treatment method and system based on multimodal data fusion of the present invention, with reference to the embodiments shown in the accompanying drawings and the reference numerals. Attached Figure Description

[0034] Figure 1 A flowchart illustrating the steps of the muscle injury rehabilitation treatment method based on multimodal data fusion in this application is shown.

[0035] Figure 2 A schematic diagram of the framework of the muscle injury rehabilitation treatment system based on multimodal data fusion in this application is shown.

[0036] Figure 3 Experimental diagram showing the differences in electromyographic signals among patients with muscle injury, healthy individuals, and patients with neuritis.

[0037] Figure 4 This is a diagram showing the results of an experiment on attitude data acquisition based on the Free Exercise DB public dataset.

[0038] Figure 5 The intention is to represent the simulated scores of rehabilitation treatment for four target subjects over 1-10 days. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0040] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationships and movement of the components in a specific posture (as shown in the attached figures). If the specific posture changes, the directional indications will also change accordingly. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0041] Figure 1 A flowchart illustrating the steps of the muscle injury rehabilitation treatment method based on multimodal data fusion in this application is shown. Figure 1As shown, this application discloses a muscle injury rehabilitation treatment method based on multimodal data fusion, including the following steps: S0, locating and determining the type of muscle injury in the target object to identify the target muscle tissue; S1, collecting electromyographic (EMG) signals, electroencephalogram (EEG) signals, and posture data of the target object and preprocessing them to obtain EMG feature data, EEG index data, and posture parameter data, forming multimodal data; wherein the EMG signals are collected from the target muscle tissue; S2, performing fusion evaluation based on the multimodal data to obtain a multimodal evaluation score; S3, generating and executing an individual rehabilitation treatment plan based on the multimodal evaluation score and the target object's personalized adaptation parameters; S4, recording the target object's rehabilitation progress data and judging the rehabilitation status based on it. When the rehabilitation status is detected to reach a preset threshold, steps S1 and S2 are triggered and re-executed, and the individual rehabilitation treatment plan is updated based on the newly generated multimodal evaluation score.

[0042] Figure 2 A schematic diagram of the framework of the muscle injury rehabilitation treatment system based on multimodal data fusion in this application is shown. Figure 2 As shown, this application also discloses a system for implementing the aforementioned method, comprising: a damage diagnosis module configured to locate and determine the type of muscle damage in a target object to identify the target muscle tissue; a data acquisition and processing module configured to include: an electromyography (EMG) detection module for acquiring EMG signals from the target muscle tissue; an electroencephalogram (EEG) acquisition module for acquiring EEG signals from the target object; a posture acquisition module for acquiring the human posture of the target object; and a preprocessing module for preprocessing the EMG signals, EEG signals, and posture data to obtain EMG feature data, EEG index data, and posture parameter data, forming multimodal data. This system also includes: a fusion assessment module, which is constructed to perform fusion assessment based on multimodal data to obtain a multimodal assessment score; a plan generation module, which is constructed to generate and execute an individual rehabilitation treatment plan based on the multimodal assessment score and the personalized adaptation parameters of the target object; and a progress tracking and update management module, which records the rehabilitation progress data of the target object and judges the rehabilitation status based on it. When the rehabilitation status is detected to reach a preset threshold, the data acquisition and processing module and the fusion assessment module are triggered to work again to generate a new multimodal assessment score and update the individual rehabilitation treatment plan based on the newly generated multimodal assessment score.

[0043] In the specific implementation of the aforementioned methods and systems, the target object first undergoes damage localization and type determination via the damage diagnosis module. Specifically, after the target object enters the system, damage feature detection is performed, and damage-related information (damaged tissue type, location, degree, and functional impairment level) is entered. Based on a pre-set anatomical structure database, the system automatically matches corresponding muscle groups and associated joints to complete the sampling area localization and generate an electrode arrangement diagram and posture acquisition range markers.

[0044] Next, electromyography (EMG) signal acquisition is performed by the EMG detection module. The acquisition process employs an adaptive electrode mode switching mechanism, automatically switching between needle / surface hybrid electrode acquisition modes based on the target muscle tissue thickness and signal attenuation coefficient. Specifically, in a particular embodiment, the electrode type is adaptively recommended based on the depth of the target muscle tissue; needle electrodes are used for deep muscles, and surface electrodes for superficial muscles. A dynamic sampling range of 2000-8000Hz is simultaneously set, and the sampling frequency is calibrated in real-time through initial signal amplitude feedback, ensuring the effectiveness and stability of the acquired signal. In other words, the sampling frequency of the EMG data is dynamically adjusted between 2000Hz and 8000Hz based on the initial signal amplitude.

[0045] In a specific embodiment, the electromyography (EMG) detection module (e.g., an EMG analyzer) automatically switches electrode modes based on the localization results: needle electrodes are used for deep muscles (such as the deep triceps brachii and the medial head of the quadriceps femoris), with an electrode diameter of 0.3 mm and a puncture depth adaptively controlled by the muscle thickness (range 2-5 mm); surface electrodes are used for superficial muscles (such as the biceps brachii and gastrocnemius), with a fixed electrode spacing of 2 cm, and signal conduction is enhanced by conductive gel. The sampling frequency is initially set to 4000 Hz and dynamically adjusted based on the amplitude of the first frame signal (threshold range 5-50 μV). When the amplitude is below 5 μV, the frequency is increased to 8000 Hz, and when it is above 50 μV, it is decreased to 2000 Hz, achieving a balance between sampling efficiency and signal quality.

[0046] The steps for hierarchical denoising and feature extraction of electromyography (EMG) signals include: eliminating power frequency interference using an adaptive bandwidth notch filter; calibrating the signal baseline using an algorithm based on the signal time-series gradient to eliminate low-frequency drift; separating the EMG signal from complex electromagnetic interference using an independent component analysis algorithm combined with dynamic sparsity constraints, and reconstructing the signal; verifying the features of the EMG signals before and after denoising to ensure the existence of key EMG features; and extracting the EMG amplitude, integral EMG, and root mean square value within a dynamic time window based on the denoised EMG signal.

[0047] Specifically, the acquired raw electromyography (EMG) signals undergo hierarchical denoising. First, a noise classification module quantifies and identifies three types of interference: power frequency interference (50 / 60Hz adaptive matching), low-frequency drift, and complex electromagnetic interference. Correspondingly, frequency band adaptive notch filtering, signal temporal gradient baseline calibration, and an improved ICA and sparse representation fusion algorithm (introducing dynamic sparse constraint coefficients) are used to achieve interference removal and signal reconstruction. Simultaneously, a feature verification module compares the integrity of key frequency bands before and after denoising to ensure that effective EMG features are not lost. Based on the denoised signal, EMG amplitude, integrated EMG (iEMG), and root mean square (RMS) parameters are extracted within a dynamic time window of 5-80ms. A lightweight feature selection algorithm retains associated feature terms and completes normalization processing. The results are then input into a preset scoring model to output an EMG activation score of 0-100 points.

[0048] In specific embodiments, regarding power frequency interference removal, a notch filter can be used to automatically identify the power grid frequency (50Hz / 60Hz), with a bandwidth of 1Hz. The target frequency signal is attenuated through digital convolution operations, and the baseline noise of the filtered signal is ≤2μV. Regarding low-frequency drift processing, a signal time-series gradient baseline calibration algorithm can be used to calculate the mean gradient of 100 consecutive frames of signal, remove drift components with an absolute gradient value >0.05μV / ms, and then calibrate the signal baseline to zero. Regarding composite interference processing, an improved ICA and sparse representation fusion algorithm can be used, with the dynamic sparsity constraint coefficient set to 0.02, to decompose the signal into 16 independent components, retain the components corresponding to electromyographic features, and reconstruct them. Regarding feature verification, the signal energy in the effective frequency band of 20-500Hz before and after denoising can be compared. An energy loss of ≤5% is considered acceptable; otherwise, the denoising process is repeated.

[0049] Feature extraction and scoring of electromyographic signals can be achieved through the following methods:

[0050] Based on a dynamic time window that adaptively adjusts according to signal complexity (5ms for simple actions, 80ms for complex actions), the electromyography (EMG) amplitude (difference between maximum and minimum values), integrated EMG (iEMG), and root mean square (RMS) values ​​are extracted. After Z-score normalization, the data are input into a logistic regression scoring model, which outputs an EMG activation score of 0-100 points. The model fit is R² ≥ 0.92. The calculation formula is as follows:

[0051] Electromyographic amplitude: A = max{|x(1)|,|x(2)|,...,|x(N)|} (1;

[0052] in (n) represents the electromyographic signal value of the nth sampling point; N represents the total number of sampling points within the dynamic time window.

[0053] Integral electromyography (iEMG) (2);

[0054] in (n) represents the electromyographic signal value of the nth sampling point; N represents the total number of sampling points within the window; , (Sampling frequency).

[0055] Root mean square value: (3);

[0056] in (n) represents the electromyographic signal value of the nth sampling point; N represents the total number of sampling points within the window.

[0057] Meanwhile, it is particularly important to emphasize that the electromyography (EMG) signal acquisition based on the EMG detection module is carried out throughout the entire treatment and assessment process. In the rehabilitation training of the target subjects, mechanical parameters such as muscle contraction strength, contraction speed, and contraction duration are collected in real time. Through a low-noise amplification circuit (with adaptively adjustable amplification factor) and secondary filtering (wavelet packet threshold denoising), a standardized set of mechanical parameters is output, which complements the statically acquired EMG characteristic parameters.

[0058] Electromyography (EMG) signal acquisition is a clinical tool for assessing muscle injury. This application processed and analyzed publicly available clinical trial datasets—Examples of Electromyograms—from PhysioNet, the open-access biomedical signaling research resource platform of the National Institutes of Health (NIH).

[0059] Experimental graphs showing the differences in electromyographic signals between patients with muscle injuries, healthy individuals, and patients with neuritis are shown below. Figure 3 As shown, the amplitude and shape of its voltage amplitude curve are significantly different from those of normal individuals and patients with neuritis, thus establishing a basis for muscle injury assessment based on electromyography (EMG) signals. Furthermore, the experimental timing sequence reveals that the EMG signals are sampled at the millisecond level, which is why the aforementioned steps employed an EEG-EMG timing synchronization calibration algorithm.

[0060] In this invention, a flexible array of surface electrodes is used to non-invasively adhere to the surface of the target muscle group, collecting parameters such as muscle contraction strength, speed, and duration in real time. Electrode impedance is monitored in real time to ensure it is ≤5kΩ, guaranteeing signal quality. A low-noise amplifier circuit adaptively adjusts the amplification factor by 1000-10000 times based on the signal amplitude, followed by secondary filtering using a db4 wavelet base wavelet packet thresholding method. After parameter extraction, normalization is performed to output a standardized set of mechanical parameters, which complements the static electromyographic characteristic parameters.

[0061] Regarding EEG signal acquisition, it can be achieved by using a portable EEG headband mounted on the EEG acquisition module to acquire the target's brainwave signals. Specifically, an EEG-EMG timing synchronization calibration algorithm is used to assess two indicators in real time: attention concentration and pain correlation index. This acquisition and assessment process runs through the entire rehabilitation assessment and treatment process, providing neuropsychological data support for multimodal fusion.

[0062] In a specific embodiment regarding EEG signal acquisition, the target subject wears a portable EEG headband with electrodes attached to the forehead area. The headband automatically detects the electrode contact impedance (impedance ≤ 5kΩ is considered acceptable) and transmits the signal to a computer via Bluetooth. The sampling period is dynamically adjusted according to the activity phase (30ms during resting phase and 10ms during training phase). An EEG-EMG timing synchronization calibration algorithm is used, with EMG signal timestamps as a benchmark, to correct for EEG signal transmission delay (delay ≤ 2ms). Two indicators are evaluated: attention concentration (based on the energy proportion of the β wave band, range 0-100) and pain association index (based on the amplitude ratio of delta waves to alpha waves, range 0-100). The evaluation results are stored in real time and synchronized to the fusion evaluation module.

[0063] Regarding the collection of posture data, the target object is simultaneously guided to the designated collection area. The posture acquisition module uses a high-definition camera as the core device, integrating the Google MediaPipe Pose open-source framework and Google technical interfaces that have undergone secondary adaptation and optimization. Through the joint key point enhancement recognition algorithm (to improve recognition accuracy in occluded scenes), it captures limb posture data in real time, extracts joint key point coordinates, joint movement angles, and motion stability parameters, and uses a multi-frame data fusion calibration algorithm to calculate parameter deviation values ​​to ensure that the deviation accuracy is ≤0.5°. At the same time, abnormal posture frame data is removed to ensure the reliability of posture data.

[0064] In a specific embodiment regarding posture data acquisition, the target object stands in a designated acquisition area (2m×2m, with positioning markers on the ground). Two high-definition cameras are simultaneously activated, integrating the Google MediaPipePose open-source framework (which has undergone secondary adaptation and optimization). A joint keypoint enhancement and recognition algorithm is enabled—for occlusion scenarios (such as arms crossing and obscuring the torso), the coordinates of key points are predicted and completed through multi-frame images. The coordinates of 33 joint keypoints are captured in real time, and the joint movement angles and motion stability parameters (based on the variance of coordinates from 50 consecutive frames) are calculated. A multi-frame data fusion calibration algorithm (taking the average of 3 consecutive frames) is used to ensure that the joint angle calculation deviation is ≤0.5°. Abnormal frames with coordinate fluctuation values ​​>2cm are removed, and standardized posture data is output.

[0065] Taking a muscle training scenario from the open-source and free database Free Exercise DB as an example, this study uses the Google BlazePose GHUM 3D model, specifically the Full version. The experimental desktop computer was configured with a 12th-generation Intel i7 CPU, an RTX 4070 GPU, and 32GB of DDR4 memory. The posture data acquisition algorithm was run, and the experimental results on multiple users and in multiple scenarios were evaluated. Figure 4 As shown, the method described in this application can accurately capture human posture joints.

[0066] In a particularly preferred embodiment, the images can be further aggregated into a video stream experiment. A 60fps camera is used, and the algorithm maintains high accuracy while processing each frame at a speed of less than 20ms, which can meet the real-time acquisition requirements when the posture changes dynamically.

[0067] Step S2 regarding the multimodal data fusion evaluation includes the following steps: S21, extracting the timestamps of each multimodal data point, and mapping the non-uniformly sampled multimodal data to a preset reference time axis using an interpolation algorithm, so that each reference time point corresponds to a set of multimodal data; S22, assigning a sampling period adaptation coefficient to each of the multimodal data mapped to the reference time axis; S23, constructing a three-dimensional modal evaluation model including electromyography modality, electroencephalography modality, and posture modality, and assigning modal weights to each modality, and assigning corresponding modal weights to the corresponding modal data. In step S24, sub-weights are assigned to the feature indicators under each mode, and the sum of the sub-weights of all feature indicators under each mode is equal to the mode weight of the corresponding mode; in step S25, the single-mode evaluation value of the corresponding mode is calculated based on the corresponding data of the feature indicators under each mode and the sub-weights of the corresponding feature indicators. The obtained single-mode evaluation value is multiplied by the sampling period adaptation coefficient assigned to the corresponding mode in step S22 to perform credibility modulation; in step S26, the modal evaluation values ​​after credibility modulation are weighted and summed according to the corresponding assigned modal weights to obtain the multimodal evaluation score.

[0068] Each modality can be considered as a dimension; that is, the electromyography modality can be considered as a biomechanical dimension, the electroencephalography modality as a neuropsychological dimension, and the posture modality as a motor posture modality. This allows the three-dimensional modal assessment model to be viewed as a "biomechanical-motor posture-neuropsychological" three-dimensional assessment model. In other words, in step S2, a "biomechanical-motor posture-neuropsychological" three-dimensional assessment model is constructed.

[0069] The specific implementation process of step S2 can be as follows: In view of the non-uniform periodic sampling characteristics of electromyography detection, electroencephalography acquisition and posture acquisition, an asynchronous data temporal collaborative processing mechanism is added. The core idea of ​​"temporal alignment-dynamic adaptation-weighted fusion" is to solve the problem of asynchronous collaboration of multimodal data. Then, through the dynamic weight allocation mechanism (adaptively adjusted based on the evidence-based data of the database module and the reliability of each modality sampling), a comprehensive evaluation score of 0-100 points is calculated according to the physiological and mechanical dimension weight S1, the movement posture dimension weight S2, and the neuropsychological dimension weight S3 (S1+S2+S3=100%).

[0070] Specifically, due to differences in detection principles and data requirements, the sampling periods for different modalities exhibit varying characteristics: electromyography (EMG) detection uses a high-frequency real-time sampling mode, electroencephalography (EEG) acquisition uses a stable continuous sampling mode, and posture acquisition uses a sampling mode that dynamically adjusts according to the complexity of the movement. In other words, EMG signals, EEG signals, and posture data present asynchronous, non-uniformly periodic multimodal data.

[0071] For these asynchronous data, the processing method is as follows: First, the timestamps of each modality's data are extracted through the time-series alignment module. A combination of linear interpolation and sliding window fusion is used to uniformly map asynchronous data from different periods to a preset baseline time-series axis. Data frames with sampling delays exceeding a preset threshold are synchronously removed to ensure the consistency of the data's time dimension. Then, a sampling period adaptation coefficient K is introduced, which is dynamically assigned based on the sampling frequency and period stability of each modality. The higher the sampling frequency and the more stable the period, the closer the adaptation coefficient K is to the set peak value, and the higher the trust weight of the corresponding data, thus achieving the collaborative fusion of non-uniform period data. Next, the multimodal assessment score is calculated. The electromyography activation score of the physiological and mechanical dimension and the assessment data of the neuropsychological dimension are calculated in a coordinated manner through the electromyography-brain-computer interface fusion assessment formula. This formula embeds time-aligned data calibration items. The core of the formula is the electromyography function assessment score (including electromyography amplitude, integral electromyography, root mean square value, and contraction stability index) and the brain-computer interface neuropsychological assessment score (including attention and pain tolerance index). Combined with the corresponding modal sampling period adaptation coefficient, the fusion assessment result is obtained by weighted summation according to preset weights and serves as the key input data for the three-dimensional assessment model. The weighting of indicators within each dimension satisfies the following: In the biomechanical dimension, the sum of the weights of muscle contraction strength achievement rate (M1), electromyographic activation score (M2), and muscle contraction stability score (M3) is S1 (M1+M2+M3=S1); In the movement posture dimension, the sum of the weights of joint range of motion achievement rate (W1), movement stability score (W2), and movement trajectory deviation value (W3) is S2 (W1+W2+W3=S2); In the neuropsychological dimension, the sum of the weights of attention score (F1) and pain association index (F2) is S3 (F1+F2=S3).

[0072] In a specific embodiment, the fusion evaluation module initiates an asynchronous data timing collaborative processing mechanism. The specific process can be summarized as follows: 1) Timing alignment: Extract timestamps from EMG, EEG, and posture data, set a 10ms reference timing axis, and use a linear interpolation algorithm to map posture data (long period) and EEG data (medium period) to the reference axis, while simultaneously removing data frames with a delay > 2ms; 2) Sampling period adaptation coefficient calculation: The sampling period adaptation coefficient K ranges from 0.6 to 1.0, with K = 1.0 for EMG signals (high-frequency stable sampling) and K = 1.0 for EEG signals (stable sampling). 0.8, posture data (dynamic cycle) K=0.6-0.8 (the more stable the cycle, the closer the K value is to 0.8); 3) Three-dimensional modal fusion calculation: dynamic weight allocation is based on the database evidence-based data to set initial values ​​(S1=40%, S2=35%, S3=25%), and further subdivided according to the weight of indicators within the dimension: M1=15%, M2=15%, M3=10% (total S1=40%); W1=15%, W2=10%, W3=10% (total S2=35%); F1=10%, F2=15% (total S3=25%). The electromyography-brain-computer interface fusion evaluation formula is: fusion score = (electromyography function evaluation score × K1 + neuropsychological evaluation score × K2) × the weight ratio of S1 and S3, where K1 and K2 are the corresponding modal adaptation coefficients, and the final output is a comprehensive evaluation score of 0-100.

[0073] The specific calculation formulas involved are as follows:

[0074] Linear interpolation algorithm: (4);

[0075] in This represents the signal value at the reference time point after interpolation; Indicates the original data collection time point The corresponding signal values ​​(such as joint angles and EEG amplitudes) are selected from the valid raw data and compared with the baseline time point. The two nearest time points and (satisfy ).

[0076] Regarding step S3, it can be achieved through the following steps: S31, based on the multimodal assessment score, a target rehabilitation movement is matched from a pre-stored standard movement library containing multiple standard rehabilitation training movements and their corresponding biomechanical parameters; S32, based on the standard movement parameters of the target rehabilitation movement and combined with the individual parameters of the target object, at least one set of initial training parameters is constructed to form an initial individual rehabilitation treatment plan; S33, during the execution of the treatment plan, electromyographic signals, electroencephalographic signals, and posture data of the target object are collected in real time, and the real-time data is compared and verified with the standard movement parameters and the expected performance values ​​set based on the initial training parameters; and the initial training parameters are dynamically adjusted according to the comparison results to optimize the individual rehabilitation treatment plan.

[0077] Step S33, which involves dynamically adjusting the initial training parameters based on the comparison results, includes: when real-time comparison and verification reveals that at least one of the following—muscle contraction strength, joint range of motion, or movement stability—failes to meet the expected performance value, providing text, voice, or visual guidance through real-time feedback, and adjusting at least one of the following initial training parameters—training intensity, movement difficulty, or training duration.

[0078] The personalized adaptation parameters for the target object in step S3 are obtained through the following steps: The target object's postural parameters and injury characteristics are collected and compared with evidence-based rehabilitation data from a population database. The training target parameters are dynamically calibrated using an individual-population data comparison algorithm to generate a training threshold range for the target object, which serves as the personalized adaptation parameter. Specifically, a personalized adaptation module can be used to collect the target object's postural parameters (height, weight, limb length) and injury characteristics (type of damaged tissue, location of damage, degree of damage, level of functional impairment). Through an individual-population data comparison algorithm, combined with evidence-based rehabilitation data from a population database module, the training target parameters are dynamically calibrated and personalized to generate a dedicated training threshold range for the target object.

[0079] In one specific embodiment, the personalized adaptation module collects the target object's body parameters (height, weight, limb length) with an accuracy of ≤1cm using an infrared rangefinder. Combined with the recorded injury characteristic information, it calls up the group evidence-based data (stored according to injury type and body characteristics) in the database module, uses an individual-group data comparison algorithm to calculate the deviation between the target object and the group mean, dynamically calibrates the training target parameters—such as muscle contraction force threshold (adjusted based on the group mean ±10%) and joint range of motion (adapted based on limb length), generates a unique training threshold range, and stores it in the target object's personalized database.

[0080] The calculation formulas involved are as follows: (5);

[0081] Among them, the height deviation rate: Limb length deviation rate: .

[0082] It should be noted that, in this application, the database module can be used to classify and store the target object's personalized parameters, real-time training data, rehabilitation progress curves, multimodal assessment results, group evidence-based data, and standard action library data, supporting real-time data retrieval and updates.

[0083] The treatment plan can be generated using the plan generation module. Specifically, based on multimodal assessment scores, the module uses an action matching algorithm to call the standard action library in the database module. This standard action library stores various rehabilitation training standard action data categorized by rehabilitation scenario, covering multiple scenarios such as bone injury, ligament injury, cartilage injury, and skeletal muscle injury. It clearly defines the target rehabilitation tissue, key joint range of motion, action timing parameters, and force control corresponding to each action. Simultaneously, it dynamically adjusts the training intensity, action difficulty, training duration, and training frequency according to personalized adaptation parameters to generate a personalized treatment plan, and supports receiving real-time feedback data to achieve dynamic optimization of the plan. During the specific implementation phase of the treatment plan, the electromyography (EMG) detection module, electroencephalography (EEG) acquisition module, and posture acquisition module are activated simultaneously to build a real-time comparison and verification mechanism: using the target action parameters in the standard action library as a benchmark, the collected EMG signals, EEG signals, and limb posture data of the target object are compared with the benchmark parameters in real time to verify whether the muscle contraction strength, contraction stability, neuropsychological state, and limb posture meet the preset treatment target values. During the treatment process, the target's response is captured throughout using the aforementioned three-dimensional monitoring methods. The degree of achievement of the treatment plan is determined by combining the dynamic assessment results output in real time from three dimensions: biomechanics (muscle contraction state), movement posture (joint activity, movement trajectory), and neuropsychology (attention, pain correlation index). The deviation between the current state and the preset target value is identified. For any deviations that do not meet the target, feedback is sent to the treatment plan generation module in real time to dynamically fine-tune the training parameters and ensure that the treatment process matches the target value.

[0084] In a specific embodiment, the treatment plan generation module, based on the comprehensive evaluation score, calls the standard action library through an action matching algorithm. The standard action library is classified according to injury scenarios (bone, ligament, cartilage, skeletal muscle injury). Each action includes the target rehabilitation tissue, the key joint range of motion (such as knee joint flexion of 0-90°), the action timing parameters (such as uniform contraction for 2s and relaxation for 1s), and the key points of force control (such as dominant force exerted by the quadriceps femoris). The training parameters are adjusted according to the personalized adaptation parameters: for mild injuries (comprehensive score of 60-80 points), set low intensity, low difficulty, and short duration (20 minutes each time, frequency of once a day). For moderate injuries (comprehensive score of 40-60 points), set medium intensity, medium difficulty, and medium duration (30 minutes each time, frequency of once a day). For severe injuries (comprehensive score < 40 points), set low intensity, low difficulty, and short duration (15 minutes each time, frequency of once every other day).

[0085] The experimental data of the 10-day rehabilitation training of 4 target objects are shown in Table 1 below and the appendix Figure 5 as follows:

[0086] Table 1 Basic Information and Training Parameter Table

[0087]

[0088] In this application, during the treatment implementation stage, the collection of electromyogram, electroencephalogram, and posture can be started simultaneously, and a real-time comparison and verification mechanism is constructed: based on the standard action parameters, the electromyogram detector compares the muscle contraction strength, stability, and threshold deviation in real time, the electroencephalogram headband monitors the change in attention, and the posture acquisition system compares the joint angle and motion trajectory deviation. The fusion evaluation module outputs a dynamic evaluation result every 100 ms to determine the achievement degree - the achievement degree ≥ 90% is qualified, 60%-90% is basically qualified, and < 60% is unqualified. For unqualified items (such as joint angle deviation > 5° and insufficient muscle contraction strength), it is real-time fed back to the plan generation module to dynamically fine-tune the parameters (such as reducing the action difficulty and extending the contraction time).

[0089] The real-time feedback in the above steps can be realized with the help of the real-time feedback module. Specifically, the real-time feedback module adopts a multi-channel prompt mechanism, outputs parameter abnormality adjustment prompts in multiple forms such as text, voice, and visual guiding lines, provides targeted guidance for situations such as muscle contraction strength deviation and posture angle abnormality, and collects the real-time data after adjustment and feeds it back to the treatment plan generation module to form a dynamic optimization closed-loop of the treatment plan. In a specific embodiment, the text prompt is displayed on the touch screen, the voice prompt is output through the built-in speaker, the volume is set to be adjustable, and the visual guiding line is superimposed and displayed on the touch screen. For example, the red color indicates the deviation direction, the green color identifies the standard trajectory, and targeted guidance is given for different deviation items, such as "The left knee joint angle is insufficient, please bend appropriately", etc.

[0090] Furthermore, this application also includes an interactive display module, which employs a layered visualization mechanism. Based on a standard movement library, it enables a 3D demonstration of rehabilitation training movements, synchronously displaying core assessment parameters, training data, and progress curves in real time. It distinguishes between core and auxiliary indicators, improving data readability and ease of operation. In a specific embodiment, the upper layer displays a 3D standard movement demonstration, the middle layer displays core indicators such as electromyographic activation scores and joint angles, and the lower layer displays progress curves, allowing the target object to view its status in real time.

[0091] Additionally, this application includes a progress tracking module, which is constructed to record multi-dimensional data for each training session at preset time points, generate a rehabilitation progress curve using a trend fitting algorithm, and monitor changes in the rehabilitation status of the target subject in real time. When improvement in rehabilitation data is detected (with a set improvement threshold), the fusion assessment module is automatically triggered to re-execute the assessment process, inputting the new assessment results into the treatment system, thus achieving a closed loop of assessment-treatment-progress tracking. In a specific embodiment, the progress tracking module records multi-dimensional data for each training cycle, generates a rehabilitation progress curve using a polynomial fitting algorithm (order 3), sets an improvement threshold (an improvement of ≥5 points in the overall score after 3 consecutive training sessions), and automatically triggers the fusion assessment module to re-assess when the threshold is met. The new assessment results are synchronized to the treatment system, and the training plan is adjusted to achieve a closed loop of "assessment-treatment-progress tracking-reassessment". The specific calculation formula is as follows:

[0092] (6).

[0093] The list of simulated progress tracking data is shown in Table 2 below:

[0094] Table 2 Progress Tracking Data List

[0095]

[0096] All the steps S0, S1 and S2 mentioned above together constitute the rehabilitation assessment steps, realizing multimodal data collection, fusion calculation and comprehensive assessment; steps S3 to S4 constitute the rehabilitation treatment steps, realizing personalized adaptation, plan generation, real-time feedback and interactive display; the two systems are linked through the progress tracking module and combined with the database module to realize full-process data communication and closed-loop operation, ensuring the accuracy of rehabilitation assessment and the adaptability of treatment plan.

[0097] This application provides a non-invasive human monitoring method and system (including posture acquisition module, EEG signal acquisition module, EMG signal acquisition module, and related technical architecture). The description of its technical solution focuses on signal processing logic, algorithm flow, device hardware connection relationship and functional implementation principle. The experimental data used are all from open source public datasets or simulated data to clearly reveal the workflow of the method. The entire process does not violate ethical principles by conducting human trials in an unauthorized manner, nor does it use clinical data that infringes on patient privacy.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A muscle injury rehabilitation treatment method based on multimodal data fusion, characterized in that, Includes the following steps: S1. Collect electromyography (EMG) signals, electroencephalography (EEG) signals, and posture data of the target object and perform preprocessing to obtain EMG feature data, EEG index data, and posture parameter data, forming multimodal data. S2, based on multimodal data fusion evaluation, to obtain multimodal evaluation scores; S3 generates and executes individualized rehabilitation treatment plans based on multimodal assessment scores and personalized adaptation parameters of the target subject; S4: Record the rehabilitation progress data of the target object and judge the rehabilitation status based on it. When the rehabilitation status is detected to reach the preset threshold, trigger and re-execute steps S1 and S2, and update the individual rehabilitation treatment plan based on the newly generated multimodal assessment score.

2. The method according to claim 1, characterized in that, Step S2 includes: S21, extract the timestamps of each of the multimodal data, and map the non-uniformly sampled multimodal data to a preset reference time axis through an interpolation algorithm, so that each reference time point can correspond to a set of multimodal data; S22, assign a sampling period adaptation coefficient to the multimodal data mapped to the reference time axis; S23, construct a three-dimensional modal assessment model including electromyography modality, electroencephalography modality and posture modality, and assign modal weights to each modality and assign sub-weights to the feature indicators under the corresponding modality. The sum of the sub-weights of all feature indicators under each modality is equal to the modal weight of the corresponding modality. S24. Based on the corresponding data of the feature indicators under each mode and the sub-weights of the corresponding feature indicators, the single-mode evaluation value of the corresponding mode is calculated respectively. The obtained single-mode evaluation value is multiplied with the sampling period adaptation coefficient assigned to the corresponding mode in step S22 to perform credibility modulation. S25, the modal evaluation values ​​after confidence modulation are weighted and summed according to the corresponding assigned modal weights to obtain the multimodal evaluation score.

3. The method according to claim 2, characterized in that, The characteristic indicators of the electromyography modality include the muscle contraction strength achievement rate, electromyography activation score, and muscle contraction stability score; the characteristic indicators of the electroencephalography modality include the attention score and pain association index; the characteristic indicators of the posture modality include the joint range of motion achievement rate, movement stability score, and movement trajectory deviation value.

4. The method according to claim 1, characterized in that, Step S3 includes: S31, Based on the multimodal assessment score, match the target rehabilitation movement from a pre-stored standard movement library containing multiple standard rehabilitation training movements and their corresponding biomechanical parameters; S32, Based on the standard movement parameters of the target rehabilitation movement and combined with the individual parameters of the target object, construct at least one set of initial training parameters to form an initial individual rehabilitation treatment plan; S33, during the execution of the treatment plan, real-time electromyography (EMG), electroencephalography (EEG), and posture data of the target subject are collected. The real-time data are compared and verified with standard movement parameters and expected performance values ​​set based on initial training parameters. Based on the comparison results, the initial training parameters are dynamically adjusted to optimize the individual rehabilitation treatment plan.

5. The method according to claim 4, characterized in that, The step S33, which involves dynamically adjusting the initial training parameters based on the comparison results, includes: when real-time comparison and verification finds that at least one of the muscle contraction strength, joint range of motion, or movement stability has not reached the expected performance value, providing text, voice, or visual guidance through real-time feedback, and adjusting at least one of the initial training parameters, namely training intensity, movement difficulty, or training duration.

6. The method according to claim 1, characterized in that, The preprocessing includes hierarchical denoising and feature extraction of the electromyographic signals, assessment of attention concentration and pain correlation index of the electroencephalogram (EEG) signals, and identification of joint key points and extraction of kinematic parameters of the posture data.

7. The method according to claim 6, characterized in that, The preprocessing steps for hierarchical denoising and feature extraction of electromyographic signals include: Notch filters with adaptive bandwidth eliminate power frequency interference; Algorithms based on signal timing gradients are used to calibrate the signal baseline in order to eliminate low-frequency drift. Independent component analysis algorithm with dynamic sparse constraints is used to separate electromyographic signals from complex electromagnetic interference and then reconstruct the signals. The electromyographic signals before and after denoising are subjected to feature verification to ensure that key electromyographic features are present; Based on the denoised electromyographic signal, the electromyographic amplitude, integral electromyographic value, and root mean square value are extracted within the dynamic time window.

8. The method according to claim 1, characterized in that, The personalized adaptation parameters of the target object are obtained through the following steps: collecting the target object's postural parameters and injury characteristics information, comparing them with the group evidence-based rehabilitation data in the database, dynamically calibrating the training target parameters through an individual-group data comparison algorithm, and generating the training threshold range of the target object as the personalized adaptation parameters.

9. The method according to claim 1, characterized in that, Before step S1, there is also step S0, which locates and determines the type of muscle damage to the target object in order to identify the target muscle tissue; the electromyographic signal is collected from the target muscle tissue.

10. A system for implementing the method according to any one of claims 1-9, characterized in that, include: The damage diagnosis module is designed to locate and determine the type of muscle damage in a target object in order to identify the target muscle tissue. The data acquisition and processing module is constructed as follows: The electromyography (EMG) detection module is used to acquire EMG signals from the target muscle tissue. The EEG acquisition module is used to acquire the EEG signals of the target object; The posture acquisition module is used to acquire the human posture of the target object; The preprocessing module is used to preprocess electromyography (EMG) signals, electroencephalography (EEG) signals, and posture data to obtain EMG feature data, EEG index data, and posture parameter data, forming multimodal data. The fusion evaluation module is constructed to perform fusion evaluation based on multimodal data in order to obtain a multimodal evaluation score; The treatment plan generation module is constructed to generate and execute individual rehabilitation treatment plans based on multimodal assessment scores and personalized adaptation parameters of the target object. The progress tracking and update management module records the rehabilitation progress data of the target object and judges the rehabilitation status based on it. When the rehabilitation status is detected to reach a preset threshold, the data acquisition and processing module and the fusion assessment module are triggered to work again to generate a new multimodal assessment score, and the individual rehabilitation treatment plan is updated based on the newly generated multimodal assessment score.

Images