Adaptive and robust smartphone-based step detection in multiple sclerosis

Abstract

Background: Many attempts to validate gait pipelines that process sensor data to detect gait events have focused on the detection of initial contacts only in supervised settings using a single sensor. Objective: To evaluate the performance of a gait pipeline in detecting initial/final contacts using a step detection algorithm adaptive to different test settings, smartphone wear locations, and gait impairment levels. Methods: In GaitLab (ISRCTN15993728), healthy controls (HC) and people with multiple sclerosis (PwMS; Expanded Disability Status Scale 0.0-6.5) performed supervised Two-Minute Walk Test [2MWT] (structured in-lab overground and treadmill 2MWT) during two on-site visits carrying six smartphones and unsupervised walking activities (structured and unstructured real-world walking) daily for 10-14 days using a single smartphone. Reference gait data were collected with a motion capture system or Gait Up sensors. The pipeline's performance in detecting initial/final contacts was evaluated through F1 scores and absolute temporal error with respect to reference measurement systems. Results: We studied 35 HC and 93 PwMS. Initial/final contacts were accurately detected across all smartphone wear locations. Median F1 scores for initial/final contacts on in-lab 2MWT were >=99.0%/>=97.6% in HC and >=99.0%/98.2% in PwMS. F1 scores remained high on structured (HC: 100%/100%; PwMS: 99.9%/99.5%) and unstructured real-world walking (HC: 97.8%/97.8%; PwMS: 94.4%/94.0%). Median temporal errors were <=0.08 s. Neither age, sex, disease severity, walking aid use, nor setting (outdoor/indoor) impacted pipeline performance (all p>0.05). Conclusion: This gait pipeline accurately and consistently detects initial and final contacts in PwMS across different smartphone locations and environments, highlighting its potential for real-world gait assessment.