Gait Disorder Assessment Based on a Large-Scale Clinical Trial: WiFi vs. Video vs. Doctor's Visual Inspection

TL;DR

This study compares WiFi-based, video-based, and doctor visual inspection methods for gait disorder assessment through a large-scale clinical trial, demonstrating the potential of emerging sensing modalities for medical diagnostics.

Contribution

It introduces the first large-scale validation of WiFi-based gait sensing, develops a novel video-to-RF data pipeline, and provides a comprehensive comparison with vision-based systems and neurologist assessments.

Findings

WiFi-based system achieves comparable accuracy to vision-based methods.

Large-scale clinical trial with 114 subjects validates sensing modalities.

Neurologists' visual assessment is less accurate than sensing systems.

Abstract

Neurological gait disorders affect a large population, significantly reducing life quality. This paper brings a foundational understanding to the potentials of emerging sensing modalities (e.g., WiFi) for gait disorder assessment, via conducting a one-year-long clinical trial in collaboration with the Neurology Associates of Santa Barbara. Our medical campaign encompasses 114 real subjects and a wide spectrum of disorders (e.g., Parkinson's, Neuropathy, Post Stroke, Dementia, Arthritis). We then develop the first WiFi-based gait disorder sensing system of its kind, distinguished by its scope of validation with a large and diverse patient cohort. To ensure generalizability, we mainly leverage publicly-accessible online videos of gait disorders for training, and develop a video-to-RF pipeline to convert them to synthetic RF training data. We then extensively test the system in a neurology center (i.e., the Neurology Associates of Santa Barbara). Additionally, we provide a 1-1 comparison with a vision-based system, by developing a vision-based gait assessment system under identical conditions, a first-of-its-kind comparison to our knowledge. We finally contrast both systems with neurologists' accuracy when basing evaluation solely on visual gait inspection, by designing/distributing a large survey to 70 neurologists, offering the first apples-to-apples comparison of these three sensing modalities. Our findings can help integrate these sensing systems into medical practice, working towards equitable healthcare.