Development of an Automated Approach for Objective Assessment of Fine Hand Movement Impairments

Abstract

The problem of restoring fine hand motor function remains highly relevant due to the crucial role of upper limbs in self-care, daily activities, and professional performance. Computer technologies for objective assessment and home-based rehabilitation of fine motor skills are urgently needed. The purpose of this study was to develop and evaluate an approach based on machine learning and the use of a digital pen on a graphic tablet to diagnose and, in the future, support recovery of fine motor skills in post-stroke patients. Materials and methods. We evaluated the statistical distinguishability of features differentiating healthy individuals from post-stroke patients based on pen position, height, pressure, and tilt angle on the tablet. Participants traced predefined standard patterns using a digital pen. The control group included 15 healthy volunteers, and the experimental group comprised 15 individuals with post-stroke fine motor impairment. Results. A set of features showing statistically significant differences between healthy participants and post-stroke patients was identified. Both the full dataset and a reduced dataset containing only significant features were used for binary classification. Five machine learning classifiers were compared for distinguishing between “healthy” and “patient” classes. The reduced feature set maintained high classification accuracy. Decision trees with bagging achieved 95.8% accuracy, making the proposed feature set promising for further clinical validation and screening applications.

Funding. The research was performed as part of the Development Programme of Tomsk State University of Control Systems and Radioelectronics for 2025–2036 within the Priority 2030 Strategic Academic Leadership Programme.