Stroke rehabilitation quantitative assessment method based on multi-modal sensor fusion

Abstract

The application discloses a stroke rehabilitation quantitative evaluation method based on multi-modal sensor fusion, relates to the medical rehabilitation technical field, and specifically compares the differences between patients and healthy personnel in Fugl-Meyer lower limb motor function scores, gait and balance function parameters, determines multi-modal data acquisition ranges, compares parameters collected before and after rehabilitation training, accurately screens out key objective indexes closely related to the stroke rehabilitation state, constructs a stroke rehabilitation quantitative evaluation model based on the key objective index data, verifies the constructed stroke rehabilitation quantitative evaluation model, applies the constructed model to data collected in the whole rehabilitation cycle after successful verification, realizes automatic quantitative evaluation of the recovery process, and generates a rehabilitation function quantitative evaluation report. The application effectively breaks through the technical limitations of single-mode detection through multi-modal data synchronous fusion, and can provide comprehensive, accurate and objective quantitative evaluation for post-stroke motor function recovery.

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation technology, specifically to a quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion. Background Technology

[0002] Stroke, a neurological disease with high incidence and disability rates, is showing a trend of increasing incidence and younger age of onset in my country due to the deepening of population aging and changes in lifestyle. Clinical data shows that approximately 70%-80% of stroke patients suffer from residual limb motor dysfunction, leading to loss of independent living ability, severely reducing quality of life and creating a huge social and medical burden. Restoring walking function is one of the core goals for stroke patients to return to their families and society; therefore, precise rehabilitation assessment and treatment for post-stroke walking disorders have significant clinical value.

[0003] Currently, the rehabilitation assessment and treatment methods commonly used in clinical and research institutions for post-stroke limb movement disorders mainly rely on one-on-one manual services provided by physicians, physical therapists (PTs), and occupational therapists (OTs). This traditional model has significant limitations: on the one hand, it requires a very high number of rehabilitation professionals with exceptional skills, leading to resource scarcity; on the other hand, manual assessment is inefficient, and the assessment results are highly dependent on the assessor's subjective experience, exhibiting significant individual differences and uncertainties, and lacking real-time accuracy.

[0004] Crucially, existing assessment systems typically focus only on the external kinematics of the limbs (such as joint angles and gait parameters), failing to delve into the activation states of relevant brain regions and the mechanisms of neural remodeling during patient movement. This "superficial" assessment model makes it difficult to achieve precise regulation based on central nervous system feedback in rehabilitation treatment programs, and hinders the construction of effective dynamic rehabilitation treatment models.

[0005] To address this challenge, scholars both domestically and internationally have begun exploring the use of modal detection for rehabilitation assessment. Among these, the integrated assessment based on technologies such as ultrasound (US) imaging, functional near-infrared spectroscopy (fNIRS), electroencephalography (EEG), and surface electromyography (sEMG) has become a research hotspot.

[0006] While existing technologies have made some progress in single-modal detection, they still face many technical bottlenecks. For example, while ultrasound intention recognition technology can achieve high-resolution imaging of muscle deformation and mechanical changes, offering advantages such as being non-invasive, highly sensitive (~1 mm / s), and having a large field of view, current technologies mostly use ultrasound independently with other sensors, lacking deep fusion and dynamic decoding algorithms for multi-task scenarios (flat ground, stairs, ramps) unique to lower limb walking after stroke. Furthermore, how to establish an accurate correlation model between ultrasound-recognized muscle deformation information and central motor intention is currently lacking systematic research.

[0007] In the area of ​​brain function testing, existing studies mostly employ single fNIRS or EEG techniques to monitor brain region activation. While fNIRS can effectively detect changes in blood oxygen metabolism, its spatial resolution is limited; although EEG can directly reflect neural electrical activity, it is susceptible to interference from scalp-skin conduction, and its extraction of intention recognition features for lower limb movement (such as beta-band oscillations and sample entropy analysis) remains insufficiently accurate. A single modality cannot comprehensively and objectively reflect the information coupling mechanism between the central and peripheral nervous systems.

[0008] Furthermore, most current systems fail to effectively integrate the entire information chain from peripheral sensory feedback to central motor intention. There is a lack of technical solutions for the spatiotemporal fusion of ultrasound muscle imaging, sEMG bioelectrical signals, and functional brain imaging (fNIRS-EEG). Existing assessment systems mostly remain at the level of qualitative description or simple quantitative parameter superposition, failing to construct a closed-loop evaluation system capable of real-time capture, quantitative calculation, and prediction of motor function outcomes.

[0009] Due to the lack of objective quantitative evaluation indicators, rehabilitation therapists find it difficult to dynamically adjust training loads and plans based on the patient's central activation level and peripheral muscle control status. Existing technologies have failed to provide effective solutions for how to use "load" to induce higher levels of activation in specific brain regions and how to optimize personalized treatment methods. Based on this, this application proposes a quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion. Summary of the Invention

[0010] This invention provides a quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion. By working collaboratively with multimodal sensors, a precise rehabilitation multimodal quantitative evaluation system based on ultrasound intention recognition technology is constructed and applied to predict patient rehabilitation outcomes, provide objective, comprehensive and real-time evidence for clinical rehabilitation intervention, significantly improve the recovery efficiency of motor function after stroke, and solve the problems mentioned in the background art.

[0011] This invention provides the following technical solution: a quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion, comprising the following steps:

[0012] S1. Patients at different stages of stroke recovery were selected as research subjects. The Fugl-Meyer scale was used for quantitative assessment. Before the start of rehabilitation training, clinical scale assessment was conducted to obtain the Fugl-Meyer lower limb motor function score, gait and balance function parameters of the research subjects. Age-matched healthy individuals were selected as the control group, and their Fugl-Meyer lower limb motor function score, gait and balance function parameters were collected. The differences in the collected index data between the research subjects and the control group were compared. Pearson correlation analysis was used to analyze the correlation between the collected index data and the Fugl-Meyer lower limb motor function score, gait and balance function parameters, and to determine the multimodal data to be collected, including central nervous system activity data based on functional near-infrared spectroscopy and electroencephalography, overall motor biomechanical data based on a three-dimensional motion capture system, and peripheral muscle activity data based on surface electromyography and ultrasound imaging.

[0013] S2. After the subjects completed routine rehabilitation training tasks, the Fugl-Meyer lower limb motor function score, gait and balance function parameters were reassessed. During the multimodal data acquisition process, the ultrasound co-contraction pattern reflecting motor preparation and intention and EMG precursor signals, the three-dimensional gait and balance dynamic parameters reflecting the quality of motor execution, and the fNIRS hemodynamics and EEG rhythmic activity reflecting the basis of neural activity were collected simultaneously using time synchronization technology.

[0014] S3. Compare the differences in Fugl-Meyer lower limb motor function scores, gait, and balance function parameters collected in steps S1 and S2, and use a multivariate logistic regression model for regression analysis. With functional recovery status as the dependent variable, preliminary screening of candidate objective indicators are included as independent variables. The model analysis identifies the key objective indicator data with the highest degree of covariance with functional recovery, and redundant variables without significant association are removed. Key objective indicator data include central nervous system activity data, overall motor biomechanical data, and peripheral muscle activity data.

[0015] S4. Construct a quantitative assessment model for stroke rehabilitation based on key objective indicator data;

[0016] S5. Validate the constructed quantitative assessment model for stroke rehabilitation;

[0017] S6. Apply the quantitative assessment model for stroke rehabilitation to the data collected throughout the rehabilitation cycle to achieve automated quantitative assessment of the recovery process and generate a quantitative assessment report of rehabilitation function.

[0018] Preferably, the inclusion criteria for the study subjects in S1 are: patients with first-time stroke confirmed by head CT or MRI, and who meet the following requirements: age 40-80 years; stroke recovery period, first-time onset, Brunnstrom stage II-V; clear consciousness, GCS=15 points; MMSE≥18 points; NIHSS score ≤16 points; Functional Independent Gait Scale score <2 points; no severe aphasia or severe motor dysfunction; and who have signed informed consent.

[0019] Preferably, the peripheral muscle activity data includes morphological change data of the target muscle dynamically acquired by an ultrasound imaging device, and surface electromyographic signals of the corresponding muscle synchronously acquired by an electromyographic acquisition device; the overall movement biomechanical data captures the patient's gait and balance function parameters through three-dimensional motion capture; the nervous system data includes brain region activation intensity signals and electroencephalograms acquired by a functional near-infrared spectroscopy brain imaging device.

[0020] Preferably, the overall biomechanical data acquisition time points are the calm state before performing the predetermined functional task and 2 hours after performing the predetermined function.

[0021] Preferably, the brain region activation intensity signal is analyzed by calculating resting-state functional connectivity to assess the functional network connectivity strength between different brain regions; the electroencephalogram is used to assess the overall complexity of brain neural activity by calculating the fuzzy entropy of the time series of each brain region and taking its average value.

[0022] Preferably, the monitoring time points for the brain region activation intensity signal and electroencephalogram are: measurements taken throughout the entire process of the research subject performing the predetermined rehabilitation movements.

[0023] Preferably, the specific steps for constructing the quantitative assessment model for stroke rehabilitation are as follows:

[0024] First, the selected key objective indicator data are used as independent variables, and the functional recovery status is used as the dependent variable to form a modeling dataset.

[0025] 2. Clean and standardize the data;

[0026] Third, the selected candidate independent variables are substituted into the multivariate logistic regression model for fitting. The stepwise screening method is used to determine the final independent variables of the model. At the same time, multicollinearity test is performed to eliminate highly collinear independent variables, so as to realize the quantitative assessment model for stroke rehabilitation.

[0027] Preferably, step S5 uses an experimental method to verify the constructed quantitative assessment method for stroke rehabilitation. The specific operation is as follows:

[0028] I. Study subjects were selected according to the inclusion criteria of S1, and raw data of the study subjects during the predetermined rehabilitation training process were collected, including multimodal data of peripheral muscle activity data, overall motor biomechanical data and nervous system data.

[0029] 2. Preprocess the collected multimodal data according to the standards used in constructing the quantitative assessment model for stroke rehabilitation;

[0030] Third, the preprocessed data is substituted into the constructed quantitative assessment model for stroke rehabilitation to obtain the predicted rehabilitation assessment results.

[0031] Fourth, measure the actual indicator data of the research subjects, compare the measured data with the evaluation results, calculate the degree of agreement between the predicted results and the measured results. If the degree of agreement exceeds the set threshold, the constructed quantitative assessment model for stroke rehabilitation is deemed accurate.

[0032] Preferably, the gait and balance function parameters include the patient's walking parameters, turning motion parameters, posture transition parameters from sitting to standing parameters, and plantar pressure parameters; the detection time points are the patient's calm state before the experiment and 2 hours after the experiment.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. This quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion clarifies the scope of multimodal data collection by comparing the differences in Fugl-Meyer lower limb motor function scores, gait, and balance function parameters between patients and healthy individuals. By comparing parameter collections before and after rehabilitation training, redundant data from repeated collections is effectively avoided, and key objective indicators closely related to stroke rehabilitation status are accurately selected. This avoids interference from irrelevant indicators, provides high-quality data support for subsequent model construction, and ensures that the assessment results accurately reflect the patient's actual rehabilitation status.

[0035] 2. This quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion effectively overcomes the technical limitations of single-modal detection by synchronously fusing multimodal data, and can provide a comprehensive, accurate and objective quantitative assessment of motor function recovery after stroke.

[0036] 3. This quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion, relying on the established quantitative assessment model for stroke rehabilitation, can accurately predict the rehabilitation outcome of patients, providing a reliable objective basis for clinical diagnosis and scientific research. At the same time, based on the results of multimodal quantitative assessment, it can generate targeted personalized rehabilitation intervention suggestions, truly realizing the closed-loop regulation of "quantitative assessment of needs - precise rehabilitation treatment", and significantly improving the rehabilitation efficacy of stroke patients. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the grouping situation during the implementation of Embodiment 1 of the present invention;

[0038] Figure 2 This is a schematic diagram of the experimental method flow for Phase One;

[0039] Figure 3 This is a schematic diagram of the experimental method flow for Phase Two;

[0040] Figure 4 This invention presents a method for quantitative assessment of stroke rehabilitation based on multimodal sensor fusion. Detailed Implementation

[0041] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] This invention provides one embodiment: Please refer to Figures 1-4 A quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion includes the following steps:

[0043] S1. Patients at different stages of stroke recovery were selected as study subjects. The inclusion criteria for study subjects were: patients with first-time stroke confirmed by head CT or MRI, and who met the following requirements: age 40-80 years; stroke recovery period, first-time onset, Brunnstrom stage II-V; clear consciousness, GCS=15 points; MMSE≥18 points; NIHSS score ≤16 points; Functional Independent Gait Scale score <2 points; no severe aphasia or severe motor dysfunction; and signed informed consent.

[0044] The Fugl-Meyer Assessment Scale was used for quantitative assessment. Before the start of rehabilitation training, clinical scale assessments were conducted to obtain the Fugl-Meyer lower limb motor function score, gait, and balance function parameters of the study subjects. Gait and balance function parameters were collected using a three-dimensional gait analysis and assessment system, including walking parameters, turning motion parameters, sitting-to-standing posture transition parameters, and plantar pressure parameters. The detection time points were the patient's resting state before the experiment and 2 hours after the experiment. Age-matched healthy individuals were selected as the control group, and their Fugl-Meyer lower limb motor function scores, gait, and balance function parameters were collected. The differences in the collected index data between the study subjects and the control group were compared. Pearson correlation analysis was used to analyze the correlation between the collected index data and the Fugl-Meyer lower limb motor function score, gait, and balance function parameters, determining the multimodal data to be collected, including central nervous system activity data based on functional near-infrared spectroscopy and electroencephalography, overall motor biomechanical data based on a three-dimensional motion capture system, and peripheral muscle activity data based on surface electromyography and ultrasound imaging.

[0045] Peripheral muscle activity data includes morphological changes in target muscles dynamically acquired via ultrasound imaging, and surface electromyography (EMG) signals of corresponding muscles simultaneously acquired via EMG acquisition. Overall motor biomechanical data captures gait and balance parameters through three-dimensional motion analysis; data acquisition points are the resting state before performing the predetermined functional task and 2 hours after completion of the task. Neurological data includes brain region activation intensity signals and electroencephalograms (EEGs) acquired via functional near-infrared spectroscopy (FIR) brain imaging. Brain region activation intensity signals are analyzed by calculating resting-state functional connectivity to assess the strength of functional network connections between different brain regions; EEGs are used to assess the overall complexity of brain neural activity by calculating the fuzzy entropy of each brain region's time series and averaging the values. Monitoring of brain region activation intensity signals and EEGs is conducted throughout the entire process of the study subjects performing the predetermined rehabilitation movements.

[0046] This application clarifies the scope of multimodal data collection by comparing the differences in Fugl-Meyer lower limb motor function scores, gait, and balance function parameters between patients and healthy individuals. By comparing parameter collections before and after rehabilitation training, redundant data from repeated collections is effectively avoided, and key objective indicators closely related to stroke rehabilitation status are accurately selected. This avoids interference from irrelevant indicators, provides high-quality data support for subsequent model construction, and ensures that the assessment results accurately reflect the patient's actual rehabilitation status.

[0047] S2. After the subjects completed routine rehabilitation training tasks, their Fugl-Meyer lower limb motor function scores, gait, and balance function parameters were reassessed. During multimodal data acquisition, time synchronization technology was used to simultaneously acquire ultrasound-guided contraction patterns reflecting motor preparation and intention, EMG precursor signals; three-dimensional gait and balance dynamics parameters reflecting the quality of motor execution; and fNIRS hemodynamics and EEG rhythmic activity reflecting the basis of neural activity. This ensured the temporal consistency of multidimensional data, accurately capturing the temporal correlation between "neural activity - motor intention - motor execution," and clearly reconstructing the brain activation, motor intention generation, and... The dynamic linkage process of actual exercise execution avoids the bias in correlation analysis caused by asynchronous data collection from different dimensions, improves data correlation and completeness, and the synchronously collected multi-source data can form a complete data chain of "nerve-muscle-motor", which is convenient for mining the intrinsic correlation between various indicators, providing a more accurate basis for the selection of key objective indicators, further improving the fitting degree and predictive reliability of subsequent evaluation models, and strengthening the accuracy and objectivity of evaluation. It can simultaneously capture multi-dimensional dynamic changes in the exercise preparation stage and execution stage, avoid information omissions caused by single time point collection, more realistically reflect the functional change patterns in the patient's rehabilitation process, and provide more scientific support for rehabilitation status assessment and program adjustment.

[0048] S3. Compare the differences in Fugl-Meyer lower limb motor function scores, gait, and balance function parameters collected in steps S1 and S2, and use a multivariate logistic regression model for regression analysis. With functional recovery status as the dependent variable, preliminary screening of candidate objective indicators are included as independent variables. The model analysis identifies the key objective indicator data with the highest degree of covariance with functional recovery, and removes redundant variables with no significant correlation. Key objective indicator data include central nervous system activity data, overall motor biomechanical data, and peripheral muscle activity data. By screening key objective indicator data, efficient collection of effective data is achieved, ensuring data quality, reducing the interference of irrelevant indicators on subsequent analysis, providing high-quality and highly correlated core data support for model construction, and improving the reliability and prediction accuracy of the constructed model.

[0049] S4. Construct a quantitative assessment model for stroke rehabilitation based on key objective indicator data;

[0050] The specific steps for constructing a quantitative assessment model for stroke rehabilitation are as follows:

[0051] First, the selected key objective indicator data are used as independent variables, and the functional recovery status is used as the dependent variable to form a modeling dataset.

[0052] 2. Clean and standardize the data;

[0053] Third, the selected candidate independent variables are substituted into the multivariate logistic regression model for fitting. The stepwise screening method is used to determine the final independent variables of the model. At the same time, multicollinearity test is performed to eliminate highly collinear independent variables, so as to realize the quantitative assessment model for stroke rehabilitation.

[0054] S5. Validate the constructed quantitative assessment model for stroke rehabilitation;

[0055] The constructed quantitative assessment method for stroke rehabilitation was validated using experimental methods. The specific procedures were as follows:

[0056] I. Study subjects were selected according to the inclusion criteria of S1, and raw data of the study subjects during the predetermined rehabilitation training process were collected, including multimodal data of peripheral muscle activity data, overall motor biomechanical data and nervous system data.

[0057] 2. Preprocess the collected multimodal data according to the standards used in constructing the quantitative assessment model for stroke rehabilitation;

[0058] Third, the preprocessed data is substituted into the constructed quantitative assessment model for stroke rehabilitation to obtain the predicted rehabilitation assessment results.

[0059] Fourth, measure the actual indicator data of the research subjects, compare the measured data with the evaluation results, calculate the degree of agreement between the predicted results and the measured results. If the degree of agreement exceeds the set threshold, the constructed quantitative assessment model for stroke rehabilitation is deemed accurate.

[0060] S6. Apply the quantitative assessment model for stroke rehabilitation to the data collected throughout the rehabilitation cycle to achieve automated quantitative assessment of the recovery process and generate a quantitative assessment report of rehabilitation function.

[0061] As described above, this application achieves efficient collection of multi-dimensional data such as patient movement intention, muscle function, gait balance, and brain region activation through multi-modal sensor fusion technology, avoiding the cumbersome data collection and complex operation problems of traditional assessment methods. At the same time, redundant data is eliminated through two parameter collections, simplifying the data processing process and improving assessment efficiency. Furthermore, the multi-sensor fusion system can realize synchronous data collection and analysis, reducing the operational difficulty for clinical medical staff and facilitating its widespread application.

[0062] Example 1

[0063] Selection of research subjects

[0064] This study employed a single-center prospective cohort approach, selecting ≥40 patients diagnosed with Brunnstrom stages II-V of stroke who experienced their first stroke between October 2023 and November 2024 at the Department of Rehabilitation Medicine and outpatient clinic of the First Affiliated Hospital of the University of Science and Technology of China (Anhui Provincial Hospital). The sample size was estimated using appropriate statistical methods based on the expected effect size, significance level, and statistical power requirements (α=0.05, 1-β=0.90, and effect size δ=0.5, with n≥40 considering potential dropouts). ≥10 patients were included in each stage, and were respectively assigned to the optimized seated rehabilitation exercise experimental group and the conventional rehabilitation exercise control group. Recruitment was conducted through hospital notices, media, online platforms, and WeChat groups.

[0065] Inclusion criteria for study participants were: patients with first-time cerebral infarction confirmed by cranial CT or MRI, according to the diagnostic criteria of the "Diagnostic Criteria for Various Cerebrovascular Diseases" published by the Chinese Neuroscience Society. Participants also met the following criteria: age 40-80 years; in the recovery period after stroke, first-time onset, Brunnstrom stage II-V; clear consciousness, Glasgow Coma Scale (GCS) score = 15; Mini Mental State Examination (MMSE) score ≥ 18; National Institute of Health Stroke Scale (NIHSS) score ≤ 16; Functional Ambulation Category Scale (FAC) score < 2; no severe aphasia or severe motor dysfunction; and signed informed consent.

[0066] Exclusion criteria include: a history of two or more cerebral infarctions or cerebral hemorrhages; serious clinical complications such as thrombosis or pulmonary embolism in the active limbs; other diseases or medical history that affect motor function; severe mental illness or accompanying mental symptoms; failure to sign informed consent form; and withdrawal.

[0067] Withdrawal criteria are as follows: During the trial, a subject may withdraw from this clinical trial if any of the following conditions are met: a. The subject withdraws their informed consent and requests to withdraw from the clinical trial; b. The investigator deems it inappropriate for the subject to continue participating in this clinical trial; c. An adverse event occurs during the treatment (an adverse event form has been completed).

[0068] Adverse event management measures: risk of rebleeding or re-infarction, risk of hypoglycemia, risk of falls, risk of seizures.

[0069] During the experiment, the research subjects were divided into groups, as detailed below. Figure 1As shown.

[0070] Experimental protocol

[0071] 1. Conventional Rehabilitation Treatment Group

[0072] (1) Physical factor therapy: Low frequency pulse electrotherapy: The electrodes are placed on the rectus femoris, tibialis anterior muscle, etc., with a frequency of <100Hz and a pulse width of 0.01ms, once a day, 20min / time, for a period of 2 weeks.

[0073] Pneumatic compression therapy: Wear a pneumatic compression bag on the affected lower limb once a day for 20 minutes each time, for a period of 2 weeks.

[0074] (2) Exercise therapy: Postural therapy, turning over training, lying and sitting training, sitting and standing training, balance training, walking training, once a day, 40 minutes each time, for 2 weeks, rehabilitation training conducted by a therapist, and follow-up for 3 months.

[0075] 2. Optimized Seated Rehabilitation Exercise Experimental Group

[0076] Patients in the recovery period after stroke were seated and trained using a sitting treadmill (MOTOmed lower limb rehabilitation device). Training began after patients signed informed consent upon admission. After confirming the safety of the seat, therapists helped each participant sit down and then adjusted the armrests, pedals, and backrest according to the patient's needs, ensuring the knee flexion-extension range was 20-40°. The affected lower limb of stroke patients was immobilized with a bandage to ensure comfort and safety during rehabilitation training. All rehabilitation training tasks used a uniform training intensity. The total experimental time was 25 minutes. Two minutes of EEG were collected at rest before exercise, followed by 5 minutes of purely passive exercise, 5 minutes each of three resistance states with active assistance, and 3 minutes of EEG at rest at the end of the experiment. Marking was performed on the EEG at each time point. Later, functional near-infrared spectroscopy was combined with simultaneous data collection, along with appropriate head and neck immobilization and headgear. If necessary, straps were used to correct genu varum / valgum. This experiment was conducted once daily, 30 minutes per session, three times per week, for two weeks, with a three-month follow-up. Rehabilitation therapists provided guidance throughout the experiment.

[0077] Combined medication: No special requirements are currently in place.

[0078] The entire experimental process is divided into two stages, stage one (e.g.) Figure 2As shown in the figure): Healthy individuals of the corresponding age groups (Group 1, Group 2, Group 3, Group 4) and stroke patients with lower limb motor dysfunction (Group 5-Group 8) were all quantitatively assessed using the Fugl-Meyer scale on the first day of admission. The Fugl-Meyer Lower Limb Motor Function Scale was used to assess lower limb motor function, and the plantar pressure testing system was used to measure gait and balance function parameters. The average left-right offset of the plantar pressure center (COPD-x) was recorded to evaluate balance stability. The assessment parameters included PE (electroencephalometric fuzzy entropy), rsFC (resting functional connectivity), iEMG rectus femoris (integrated electromyography value), rectus femoris A-super muscle thickness, and serum myokinase B. Independent samples t-test and Pearson correlation analysis were used to analyze the differences between each indicator in the disease group and the control group, and the correlation between the multimodal assessment indicators and physical scores and gait indicators, respectively, to ensure that the selected multimodal indicators could accurately reflect the stroke rehabilitation status.

[0079] Phase Two (e.g.) Figure 3 (As shown in the figure): Patients with lower limb motor dysfunction after stroke (Group 5-Group 8) all received two weeks of systematic rehabilitation treatment. After the rehabilitation treatment, five physiological indicators were measured: electroencephalogram (EEG) fuzzy entropy (FE), resting-state functional connectivity (rsFC), integrated electromyography (iEMG) value of rectus femoris muscle, A-super muscle thickness of rectus femoris muscle, and serum myokinase B. Before the patients were discharged, the Fugl-Meyer Lower Limb Motor Function Scale was used to assess lower limb motor function, and the gait and balance function parameter COPD-x (average left-right displacement of the plantar pressure center) was measured to comprehensively evaluate the clinical efficacy of the rehabilitation treatment.

[0080] Evaluation indicators and testing time points

[0081] Specifically: (1) Fugl-Meyer lower limb motor function score: Scoring criteria: There are 17 activities in total for the lower limbs. 0 points are given for completely being unable to perform the activities, 1 point for being able to perform them partially, and 2 points for performing them fully. The total score for lower limb motor function assessment is 34 points. The testing time points are after admission and before discharge, with a follow-up of 1 month.

[0082] (2) Gait and balance function parameters COPD-x: scoring criteria and methods, gait analysis: collected using a three-dimensional gait analysis and assessment system. The collected parameters include the patient's walking parameters (spatial and temporal), turning motion parameters, posture transition parameters from sitting to standing, plantar pressure parameters, etc.; the detection time points are the patient's calm state before the experiment and 2 hours after the experiment. The detection time points are after admission and before discharge, and a follow-up of 1 month;

[0083] (3) Ultrasound and surface electromyography: Ultrasound and surface electromyography were used to measure the thickness of the rectus femoris muscle and iEMG (integrated electromyography value) to assess the lower limb muscle morphology and muscle activation status of stroke patients. The detection time points were after admission and before discharge.

[0084] (4) Electroencephalography (EEG): The fuzzy entropy (FE) of different brain regions was measured and the average value was taken to analyze the activation status of the brain regions. The detection time points were after admission and before discharge. The measurement was carried out throughout the entire process of optimizing the sitting rehabilitation exercise experiment.

[0085] (5) Functional near-infrared (fNIRS): measures resting functional connectivity (rsFC) and analyzes brain region activation. The detection time points are after admission and before discharge, and the entire process of optimizing seated rehabilitation exercise is measured.

[0086] Statistical processing

[0087] Statistical analysis was performed using SPSS 21.0 or R 4.0.2 software. Significance tests for baseline information were conducted using Pearson chi-square test, Fisher's exact test, Mann-Whitney test, or independent samples t-test (the appropriate statistical analysis method was selected based on the nature of the data). Multivariate logistic regression was used for multivariate analysis.

[0088] A binary logistic regression model was used, with patient rehabilitation outcome as the dependent variable (an increase in Fugl-Meyer lower limb score and a decrease in COPD-x were denoted as 1, and others as 0), and multimodal assessment indicators as independent variables. First, univariate analysis (t-test for continuous variables and chi-square test for categorical variables) was used to screen independent variables related to rehabilitation outcome (p < 0.05). Then, a multivariate logistic regression model was constructed, and the overall effectiveness of the model was verified by the likelihood ratio test (p < 0.05). The model fit was evaluated using the AIC and BIC criteria. Finally, the most relevant independent predictors of rehabilitation outcome were determined, and the prediction accuracy of rehabilitation outcome was evaluated.

[0089] Furthermore, when using the binary logistic regression analysis model, the data needs to be preprocessed. The specific operation is as follows: (1) First, the data is manually de-artifacted, that is, by browsing the EEG signal, the obviously noisy segments are cut off; then, baseline drift is removed to prevent the signal from shifting too much.

[0090] (2) Then the signal is subjected to a 0.5Hz high-pass filter and a 45Hz low-pass filter, and a 50Hz notch filter is used to remove power frequency interference.

[0091] (3) Independent Component Analysis (ICA) is used to remove electrooculogram artifacts. The basic assumption of ICA is that the electroencephalogram (EEG) signal and the electrooculogram (EOG) signal are independent of each other. The collected mixed signal can be inversely processed to obtain each independent signal component. The artifact components can be found from these components and then removed manually.

[0092] In summary, this quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion constructs a precise rehabilitation strategy quantification and clinical efficacy evaluation model based on continuous dynamic detection of stroke motor processes using ultrasound intention recognition and brain function detection technologies, and multi-sensor fusion. This overcomes the limitations of single-modality approaches, providing a comprehensive and accurate assessment of post-stroke motor function recovery. The established multi-sensor fusion quantitative assessment system is applied to predict patient rehabilitation outcomes, providing objective evidence for clinical practice and research. Furthermore, based on the multimodal quantitative assessment results, personalized rehabilitation intervention suggestions can be generated, achieving a closed-loop regulation of "quantitative needs assessment - precise rehabilitation treatment," thereby improving the rehabilitation efficacy for stroke patients.

[0093] All standard parts used in this invention are commercially available products, and irregularly shaped parts can be customized according to the specifications and drawings. All specific connection methods of the structures adopt well-known and mature technologies in the art, such as bolt connections. The machinery, parts, and equipment used are all existing models under current technical conditions. The material, size, and specifications of each component can be selected according to actual needs, and this specification does not impose any limitations on this. Content not described in detail in this specification belongs to prior art known to those skilled in the art. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention. The scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion, characterized in that: Includes the following steps: S1. Patients at different stages of stroke recovery were selected as research subjects. Before the start of rehabilitation training, clinical scale assessments were conducted to obtain the Fugl-Meyer lower limb motor function score, gait, and balance function parameters of the research subjects. Age-matched healthy individuals were selected as the control group, and the same parameters were collected. The differences in the collected index data between the research subjects and the control group were compared. Pearson correlation analysis was used to analyze the correlation between the collected index data and the Fugl-Meyer lower limb motor function score, gait, and balance function parameters, and to determine the multimodal data to be collected, including central nervous system activity data based on functional near-infrared spectroscopy and electroencephalography, overall motor biomechanical data based on a three-dimensional motion capture system, and peripheral muscle activity data based on surface electromyography and ultrasound imaging. S2. After the subjects completed routine rehabilitation training tasks, the Fugl-Meyer lower limb motor function score, gait and balance function parameters were reassessed. During the multimodal data acquisition process, the ultrasound co-contraction pattern reflecting motor preparation and intention and EMG precursor signals, the three-dimensional gait and balance dynamic parameters reflecting the quality of motor execution, and the fNIRS hemodynamics and EEG rhythmic activity reflecting the basis of neural activity were collected simultaneously using time synchronization technology. S3. Compare the differences between the data collected in S1 and S2, and use a multivariate logistic regression model for regression analysis. With functional recovery status as the dependent variable, the candidate objective indicators that have been preliminarily screened are included as independent variables. The key objective indicator data with the highest degree of covariance with functional recovery are identified through model analysis. S4. Based on key objective indicator data, construct a quantitative assessment model for stroke rehabilitation; validate the constructed quantitative assessment model for stroke rehabilitation; after successful validation, apply the constructed model to the data collected throughout the rehabilitation cycle to achieve automated quantitative assessment of its recovery process and generate a quantitative assessment report of rehabilitation function.

2. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 1, characterized in that: The inclusion criteria for the study subjects in S1 were: patients with first-time cerebral infarction confirmed by cranial CT or MRI, who met the following criteria: age 40-80 years; in the recovery period of stroke, first-time onset, Brunnstrom stage II-V; clear consciousness, GCS=15; MMSE≥18; NIHSS score ≤16; Functional Independent Gait Scale score <2; no severe aphasia or severe motor dysfunction; and signed informed consent.

3. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 1, characterized in that: The peripheral muscle activity data includes morphological change data of the target muscle dynamically acquired by an ultrasound imaging device, and surface electromyographic signals of the corresponding muscle synchronously acquired by an electromyography acquisition device; the overall movement biomechanical data captures the patient's gait and balance function parameters through three-dimensional motion capture; the nervous system data includes brain region activation intensity signals and electroencephalograms acquired by a functional near-infrared spectroscopy brain imaging device.

4. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 3, characterized in that: The overall biomechanical data acquisition time points are the calm state before performing the predetermined function task and 2 hours after the predetermined function task is performed.

5. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 3, characterized in that: The brain region activation intensity signal is analyzed by calculating resting-state functional connectivity to analyze the functional network connectivity strength between different brain regions; the electroencephalogram is used to assess the overall complexity of brain neural activity by calculating the fuzzy entropy of the time series of each brain region and taking its average value.

6. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 5, characterized in that: The monitoring time points for the brain region activation intensity signal and electroencephalogram were: measurements were taken throughout the entire process of the study subjects performing the predetermined rehabilitation movements.

7. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 1, characterized in that: The specific steps for constructing a quantitative assessment model for stroke rehabilitation are as follows: First, the selected key objective indicator data are used as independent variables, and the functional recovery status is used as the dependent variable to form a modeling dataset.

2. Clean and standardize the data; Third, the selected candidate independent variables are substituted into the multivariate logistic regression model for fitting. The stepwise screening method is used to determine the final independent variables of the model. At the same time, multicollinearity test is performed to eliminate highly collinear independent variables, so as to realize the quantitative assessment model for stroke rehabilitation.

8. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 1, characterized in that: The constructed quantitative assessment method for stroke rehabilitation was validated using experimental methods. The specific procedures were as follows: I. Study subjects were selected according to the inclusion criteria of S1, and raw data of the study subjects during the predetermined rehabilitation training process were collected, including multimodal data of peripheral muscle activity data, overall motor biomechanical data and nervous system data.

2. Preprocess the collected multimodal data according to the standards used in constructing the quantitative assessment model for stroke rehabilitation; Third, the preprocessed data is substituted into the constructed quantitative assessment model for stroke rehabilitation to obtain the predicted rehabilitation assessment results. Fourth, measure the actual indicator data of the research subjects, compare the measured data with the evaluation results, calculate the degree of agreement between the predicted results and the measured results. If the degree of agreement exceeds the set threshold, the constructed quantitative assessment model for stroke rehabilitation is deemed accurate.

9. The quantitative assessment method for stroke rehabilitation based on multimodal sensor fusion according to claim 1, characterized in that: The gait and balance function parameters include the patient's walking parameters, turning motion parameters, sitting-to-standing posture transition parameters, and plantar pressure parameters; the detection time points are the patient's calm state before the experiment and 2 hours after the experiment.

Images