To facilitate remote home- based intelligent assessment and rehabilitation guidance for upper limb motor dysfunction in stroke patients, this study aimed to explore the reliability and accuracy of a machine-learning-based intelligent assessment system for evaluating Brunnstrom stages of upper limb and hand motor function in stroke patients.

A total of 220 patients with stroke who were admitted to the Department of Rehabilitation Medicine of the First Affiliated Hospital of Sun Yat-sen University from August 1, 2024 to June 30, 2025 were enrolled. Data from 160 stroke patients were used to train and optimize the machine-learning model of the intelligent assessment system for upper-limb motor function in stroke patients. Patients completed standardized movements under video guidance and were recorded by trained therapists. Brunnstrom stages of upper limb and hand motor function were independently evaluated by therapists and the intelligent assessment system. The remaining 60 stroke patients were used to validate the system by comparing automated assessments with therapist ratings.

The machine learning-based intelligent assessment system demonstrated excellent test-retest reliability, with intraclass correlation coefficients (ICCs) of 0.989 for upper limb staging and 0.969 for hand staging. The inter-rater reliability for the intelligent assessment system was also high (upper limb ICC was 0.896; hand ICC was 0.959). Agreement between the intelligent assessment system and therapist ratings was excellent, with a Kappa coefficient of 0.900 (P<0.001) for upper limb staging and 0.816 (P<0.001) for hand staging.

The machine-learning-based intelligent assessment system demonstrated good reliability and accuracy in assessing upper limb motor function after stroke and may serve as a practical assessment tool for remote upper limb function evaluation and monitoring in patients with stroke.