ABSTRACT
Background:
Human activity recognition (HAR) technology has been advanced with the development of wearable devices and the machine learning (ML) algorithm. Although previous researches have shown the feasibility of HAR technology for home rehabilitation, there has not been enough evidence based on clinical trial.
Objective:
We intended to achieve two goals: (1) To develop a home-based rehabilitation (HBR) system, which can figure out the home rehabilitation exercise of patient based on ML algorithm and smartwatch; (2) To evaluate clinical outcomes for patients with chronic stroke using the HBR system.
Methods:
We used off-the-shelf smartwatch and the convolution neural network (CNN) of ML algorithm for developing our HBR system. It was designed to be able to share the time data of home exercise of individual patient with physical therapist. To figure out the most accurate way for detecting exercise of chronic stroke patients, we compared accuracy results with dataset of personal/total data and accelerometer only/gyroscope/accelerometer combined with gyroscope data. Using the system, we conducted a preliminary study with two groups of stroke survivors (22 participants in HBR group and 10 participants in a control group). The exercise compliance was periodically checked by phone calls in both groups. To measure clinical outcomes, we assessed the Wolf motor function test (WMFT), Fugl-meyer assessment of upper extremity (FMA-UE), grip power test, Beck’s depression index and range of motion (ROM) of the shoulder joint at 0 (baseline), 6 (mid-term), 12 weeks (final) and 18 weeks(6 weeks after the final assessment without HBR system).
Results:
The ML model created by personal data(99.9%) showed greater accuracy than total data(95.8%). The movement detection accuracy was the highest in accelerometer combined with gyroscope data (99.9%) compared to gyroscope(96.0%) or accelerometer alone(98.1%). With regards to clinical outcomes, drop-out rates of control and experimental group were 4/10 (40%) and 5/22 (22%) at 12 weeks and 10/10 (100%) and 10/22 (45%) at 18 weeks, respectively. The experimental group (N=17) showed a significant improvement in WMFT score (P=.02) and ROM (P<.01). The control group (N=6) showed a significant change only in shoulder internal rotation (P=.03).
Conclusions:
This research found that the homecare system using the commercial smartwatch and ML model can facilitate the participation of home training and improve the functional score of WMFT and shoulder ROM of flexion and internal rotation for the treatment of patients with chronic stroke. We recommend our HBR system strategy as an innovative and cost-effective homecare treatment modality. Clinical Trial: Preliminary study (Phase I)