Unsupervised Gait Assessments of Stroke Patients Using a Smartphone and Machine Learning.

Home-based rehabilitation is a trend of post-stroke lower limb rehabilitation, aimed at a long-term and higher dose of therapy. Unsupervised gait assessments can help therapists to track patients' recovery progress and timely adjust rehabilitation interventions. This study aims to develop a smartphone-based wireless system for unsupervised gait assessments of stroke patients. The proposed system is based on smartphone motion sensors and uses machine learning approaches to interpret the gait features. We characterized the ability of the proposed system to extract gait features and detect abnormal gait patterns from 9 stroke patients and 10 healthy subjects. Results showed that the proposed system demonstrated comparable performance to the Vicon motion capture system for gait feature extraction (R2 = 0.99), and that extracted gait features could be used to detect patients' abnormal gait patterns (Average accuracy = 100%). Further analysis also demonstrated the correlation between gait features and the FMA-LE score for stroke patients. We conclude that the proposed smartphone-based system has sufficient potential for unsupervised gait assessments of stroke patients.