Exoskeleton-Assisted Anthropomorphic Movement Training For The Upper Limb After Stroke:Human-Robot Interaction Mechanism And Rehabilitation Application

Objective: To calculate multidimensional kinematic parameters based on feature extraction of the high-dimensional data from motion sensors.Moreover,the study aimed to explore the associations between kinematics with clinical scales,the human-robot interaction mechanism of exoskeleton-assisted anthropomorphic movement training(EAMT),and its effectiveness for the upper limb after stroke.Methods:(1)In this study,motion sensors were used to calculate upper-limb kinematic metrics for the participants.Clinical assessments for stroke participants included the Fugl-Meyer Assessment for Upper Extremity(FMA-UE),Action Research Arm Test(ARAT),and Modified Barthel Index(MBI).Pearson correlation coefficients were conducted between the kinematic variables and clinical assessments.Multivariable linear and principal component analysis(PCA)regression models were used to account for the clinical scores of the stroke individuals.(2)Upper limb kinematic data of the patients with subacute stroke were extracted during EAMT training.The R software was used for mediation analysis of mechanical error on motor impairment and kinematic performance.The temporal trend of kinematic performance and the Fugl-Meyer Assessment for Upper Extremity(FMA-UE)was analyzed by one-way repeated measured ANOVA.(3)Twenty patients with subacute stroke were enrolled in the pilot study.Primary outcome was feasibility analysis.Secondary outcomes included the FMA-UE,ARAT,MBI and kinematic variables.In the non-inferior,two-center randomized controlled trial,80 patients with subacute stroke were enrolled.The primary measure was the change in FMA-UE at 4 weeks.Secondary outcomes were the MBI,World Health Organization Disability Assessment Schedule 2(WHODAS 2.0),kinematics,and safety assessment.Results:(1)Among the kinematic variables,mean speed,peak speed and number of motor units were correlated with the clinical scales(P<0.05).Multivariable regression analysis indicated that the only significant predictor to the clinical measures was mean velocity,accounting for 55%,51%,and 32% of the FMA-UE,ARAT,and MBI variance,respectively.The first three PCs explained 91.3% variance of the dataset for the stroke survivors.High-dimensional feature extraction further improved the variance explaining upper limb motor function.(2)Motor impairment was significantly associated with kinematic performance in the stroke patients.Motor learning process of the anthropomorphic movement included not only the direct drive of the robotic arm,but also the human-robot interaction on movement smoothness and postural error,with the indirect mediation effects of-0.241 and-0.362.Moreover,there were significant improvements on kinematics and the FMA-UE with time.(3)Pilot study showed that EAMT therapy was safe and feasible in stroke rehabilitation,which contributed to the improvement of upper limb impairment(FMA-UE difference,4.3 points),activities of daily living(MBI difference,8.7 points)and kinematics,but had no significant improvement in functional capacity(ARAT difference,1.03 points).Such results were supported by the full-power trial that EAMT presented significant improvement in motor impairment(FMA-UE difference,4.51 points),ADL(MBI difference,7.00 points),and kinematics,but no significant difference was noted regarding social participation.Conclusions: Sensor-based evaluation is objective to understand motor performance and high-dimensional feature extraction may add value to comprehensively interpret motor function for the upper limb after stroke.The exoskeleton could improve kinematic performance of the patients by adjusting the human-robot interaction force dynamically based on motor impairment.EAMT therapy appears to be safe and effective for upper limb motor impairment,activities of daily living,and kinematic performance in patients with subacute stroke through repetitive practice of human-like movements.However,it warrants further investigations into long-term effects,potential neural mechanisms,and paradigm optimization such as number of anthropomorphic movements,human-root interaction strategy,and mechanical structure.Part Ⅰ: Construction of Kinematic Parameters and Feature Extraction of High-dimensional Data for the Upper Limb after StrokeObjective: To calculate multidimensional kinematic parameters using motion sensors,and explore whether high-dimensional kinematic features could account for upper limb motor function comprehensively in stroke survivors.Methods: Thirty-seven individuals with stroke and twenty healthy adults were included in this study from December 2019 to January 2021.Clinical assessments for the stroke participants included the Fugl-Meyer Assessment for Upper Extremity(FMA-UE),Action Research Arm Test(ARAT),and Modified Barthel Index(MBI).Motion sensors were used to collect multiple degree-of-freedom data for calculating kinematic metrics,including movement time,mean velocity,peak velocity,percentage of time to peak velocity,number of motion units,and normalized integrated jerk.Pearson correlation coefficients were conducted between the kinematic variables and clinical assessments.Multivariable linear and principal component analysis(PCA)regression models were used to account for the clinical assessments in stroke patients.A two-sided P<0.05 was set as statistical significance.Results: The kinematic parameters were significantly different between stroke survivors and healthy adults(P<0.001)except the percentage of time to peak velocity.Mean speed,peak speed and number of motor units were correlated with the clinical scales(P<0.05).Multicollinearity was found between movement time and normalized integrated jerk(r=0.71,P<0.05),as well as among mean velocity,peak velocity(r=0.96,P<0.05)and number of motion units(r=-0.74,P<0.05).Multivariable regression analysis indicated that the only significant predictor to the clinical measures was mean velocity,accounting for 55%,51%,and 32% of the FMA-UE,ARAT,and MBI variance,respectively.The first three PCs explained 91.3% variance of the kinematic dataset and were associated with the clinical assessments for the stroke survivors.PCA regression models further improved the variance explaining upper limb motor function.Moreover,the demographics,including age,gender,type of stroke,and paretic side,showed no significant influence in any regression model(P>0.05).Conclusions: Sensor-based kinematic evaluation is objective to understand motor performance for the upper limb after stroke.PCA,as a high-dimensional feature extraction method,may add value to comprehensively interpret motor function after stroke.Part Ⅱ: Human-Robot Interaction Mechanism of Exoskeleton-assisted Anthropomorphic Movement Training for the Upper Limb after StrokeObjective: To explore the human-robot interaction mechanism of exoskeleton-assisted anthropomorphic movement training(EAMT)for the upper limb after stroke.Methods: In this study,kinematic data of 16 patients with subacute stroke were extracted during EAMT training for the upper limb.Pearson coefficient was used for correlation analysis.Based on the non-parametric bootstrap method,the Lavaan Package of R software was used for mediation effect analysis of mechanical error on motor impairment and kinematic performance(movement time,smoothness,postural error).Repeated measured ANOVA was conducted to analyze the temporal trend of kinematics and Fugl-Meyer Assessment for Upper Extremity(FMA-UE)at baseline,2 and 4 weeks.Statistical significance was set at 0.05.When statistically significant,multiple pairwise post-hoc comparisons were performed with the Bonferroni adjustments.Results: Upper limb motor impairment were significantly associated with the kinematic variables(r=-0.60 to-0.31,P<0.05)in the stroke individuals.Human-robot interaction had no significant effect on movement time(P=0.998).Patients’ motor learning process of the human-like movement included not only the direct drive of the exoskeletal arm,but also the human-robot interaction on movement smoothness and postural error.The indirect mediation effects were-0.241(95% Bootstrap CI,-0.437 to-0.045;40.3%,P=0.016)and-0.362(95% Bootstrap CI,-0.568 to-0.156;62.3%,P=0.001),respectively.Moreover,there were significant improvements on the kinematics and FMA-UE in stroke individuals with the 4-week EAMT therapy.Conclusions: Robot-based kinematic parameters can be used to measure upper limb motor function for the patients after stroke.The exoskeleton can improve patients’ kinematics by adjusting human-robot interaction force dynamically based on their motor impairment.EAMT therapy may be beneficial to promote motor learning of the human-like movements and facilitate upper limb motor recovery after stroke.Part Ⅲ: Effects of Exoskeleton-assisted Anthropomorphic Movement Training for the Upper Limb after StrokeObjective: To investigate the safety,feasibility,and effects of exoskeleton-assisted anthropomorphic movement training(EAMT)on upper limb motor impairment,activities of daily living,social participation,and kinematic performance for the patients with subacute stroke.Methods: This study is composed of two clinical trials.The pilot study was a single-blind,single-center,randomized controlled trial.Twenty patients with sub-acute stroke were enrolled in the Department of Rehabilitation Medicine,Tongji Hospital,Wuhan from December 2018 to May 2020.Outcome measures were conducted at baseline and after 4 weeks of intervention.Primary outcome was feasibility analysis.Secondary outcomes included the Fugl-Meyer Assessment for Upper Extremity(FMA-UE),Action Research Arm Test(ARAT),Modified Barthel Index(MBI)and kinematic variables(movement time,smoothness,and postural error).In the non-inferior,two-center randomized controlled trial,80 patients with subacute stroke were enrolled in the rehabilitation medicine Department of Tongji Hospital,and Hubei Integrated Traditional Chinese and Western Medicine Hospital,Wuhan from June 2020 to August 2021.Outcomes were assessed at baseline,after 2 and 4 weeks of intervention,with the primary measure being the change in FMA-UE at 4 weeks.Secondary outcomes were the MBI,World Health Organization Disability Assessment Schedule 2(WHODAS 2.0),kinematics,and safety assessment.Statistical significance was set at 0.05.Results: Pilot study showed that the 4-week EAMT therapy was safe and feasible for upper limb rehabilitation after stroke since no patient reported difficulty in completing the training nor did any serious adverse events occur.Compared with the conventional therapy,EAMT contributed to the improvement of upper limb impairment(FMA-UE difference,4.3 points),activities of daily living(MBI difference,8.7 points)and kinematic parameters,but had no significant improvement in functional capacity(ARAT difference,1.03 points).Such results were analogous in the full-power trial that EAMT presented significant improvement in motor impairment(FMA-UE difference,4.51 points),ADL(MBI difference,7.00 points),and kinematics,but no significant between-group difference was noted regarding social participation.Conclusions: EAMT therapy appears to be safe and effective for the patients with subacute stroke through repetitive practice of human-like movements.The results indicate preliminary evidence of EAMT therapy to improve upper limb motor impairment,activities of daily living and kinematic performance.However,it warrants further investigations into long-term effects,potential neural mechanisms,and paradigm optimization.