Remote Assessment Of Multiple Sclerosis Ambulation Using Smartphone And Smartwatch Data.pdf

This study investigates the feasibility of using smartphone and smartwatch sensors to remotely assess physical function in people with multiple sclerosis (PwMS) through the Two-Minute Walk Test (2MWT). It aims to distinguish between PwMS subgroups (mild and moderate disability) and healthy controls (HC) using machine learning (ML) techniques.

Key Findings:

- Feasibility: The study successfully collected data from 97 participants (24 HC, 52 mild disabled PwMS, 21 moderately disabled PwMS) over a 24-week period using smartphone and smartwatch sensors during daily 2MWTs.

- Feature Extraction: A large number of signal-based features related to movement were extracted from the devices' sensors, with many showing strong correlations (R>0.3) with clinical outcomes.

- Classification Performance: Using feature selection techniques (LASSO), ML models (Logistic Regression, Support Vector Machines, Random Forest) were trained and evaluated. Smartphone-based models achieved the highest classification accuracy for distinguishing PwMS from HC and mild from moderate disability PwMS.

- Subject Specificity: The study observed low intra-subject variability and high inter-subject variability in gait characteristics, indicating that each subject exhibits unique gait patterns.

- Conclusion: Consumer technology like smartphones and smartwatches can effectively capture and model ambulatory characteristics of PwMS, enabling remote and objective assessment of physical function.