An Assessment System for Post-Stroke Manual Dexterity Using Principal Component Analysis and Logistic Regression.

Hand function assessment is crucial for patients with stroke, who must perform regular repetitive tasks during rehabilitation. However, the conventional evaluation method is subjective and not uniform among physicians. A novel method is proposed in this paper to analyze raw data from a data glove equipped with 16 six-axis inertial measurement units. The proposed method can provide accurate assistance to physicians and objectively assess patients' hand function. Three tasks (the thumb task, the grip task, and the card-turning task) were conducted to evaluate participants' hand function. Representative parameters of hand function in each task and overall evaluation were extracted through principal component analysis and used to develop logistic regression models. The results revealed that all three tasks can be used to perfectly predict healthy subjects and subjects with stroke, with the thumb task exhibiting the highest predictive accuracy for the severity of hand dysfunction. Overall, the proposed method can serve as an efficient method for physicians to assess the hand function of patients with stroke.