An Intelligent Decision Support System for Stroke Rehabilitation Assessment

Assessment of rehabilitation exercises is important to determine an adequate rehabilitation intervention. However, this assessment is costly as it requires the presence of a therapist. This paper presents an interactive multimodal approach that automatically assesses exercise performance, identifies salient features of assessment, and iteratively present user-specific analysis to support therapist's decision making on personalized rehabilitation assessment. This approach can achieve good agreement level with therapist's evaluation (i.e. an average of 0.8565 F1-scores on three upper-limb exercises), which is higher than non-interactive, unimodal models.