Prediction of stroke-related diagnostic and prognostic measures using robot-based evaluation

Traditional clinical scores for assessment of impairments resulting from stroke are inherently subjective and limited by inter-rater and intra-rater reliability. In contrast, robotic technologies provide objective, highly repeatable tools for quantification of motor performance of stroke subjects. Although use of robotic technologies has been widely suggested in the literature, they are not an established tool and their relationship to traditional clinical scales for stroke diagnosis and prognosis is mostly unknown. In this study we propose the application of two non-linear system identification methods, Parallel Cascade Identification and Fast Orthogonal Search, for prediction of stroke-related clinical scores using robot-based metrics. We show the suitability of these two methods for prediction of both diagnostic and prognostic scores. We compare our results with a previously applied approach based on linear regression and show the superiority of our modeling approach. Our results also underscore the importance of quantifying proprioceptive deficits in the prediction of motor-related prognosis scores.