Wearable Sensor-Based Fine-Grained and Comprehensive Upper Extremity Motor Function Assessment for Poststroke Rehabilitation.

Abstract

Objective To develop and validate a wearable sensor-based, fine-grained assessment framework for quantitative evaluation and prediction of upper extremity motor function in poststroke patients. Design Validation cohort study with modeling of clinical predictions. Setting Hospital-based rehabilitation clinic. Participants A total of 80 poststroke patients (N=80) with upper extremity motor dysfunction were recruited and underwent routine Fugl-Meyer Assessment for Upper Extremity (FMA-UE) evaluation. Upper extremity kinematic data and corresponding clinical labels were collected using a wearable motion acquisition system. Interventions This study was observational in nature and involved no therapeutic intervention. We confirm the intervention is not applicable. Main outcome measures Agreement between clinician-rated FMA-UE scores and system-predicted FMA-UE scores. Results Wearable kinematic data and corresponding clinician-rated FMA-UE scores were collected during standardized assessment procedures. Guided by clinical labels, 8 critical items were selected from the original 33-item FMA-UE. Fine-grained, action-level scoring models were constructed for the selected items using feature engineering and machine learning techniques, enabling more discriminative and fine-grained item-level scoring. A regression model based on the fine-grained item scores was subsequently developed to predict the FMA-UE total score. The predicted scores showed strong agreement with clinician-rated scores, with an R 2 of 0.950 and a Spearman correlation coefficient of 0.972. Conclusions A wearable sensor-based fine-grained assessment framework can provide objective, high-resolution evaluation and accurate prediction of upper extremity motor function after stroke. By relying on a minimal set of key assessment items, the framework reduces assessment burden while maintaining clinical consistency with standard FMA-UE scoring.