In-Home Free-Living Gait Monitoring Using a Single Lumbar IMU Reveals Nonlinear Complexity Changes During Intensive Stroke Rehabilitation

Conventional clinical assessments of stroke rehabilitation are limited to supervised environments and fail to capture continuous real-world motor behavior. This proof-of-concept pilot study (case series, n = 3) evaluated the feasibility and longitudinal sensitivity of wearable IMU-derived digital biomarkers of three individuals with chronic stroke across Chapman University’s Stroke Boot Camp (SBC), a two-week intensive interdisciplinary rehabilitation program. A lower-back-mounted DynaPort McRoberts IMU provided continuous multi-day free-living monitoring across five sessions that spanned one week pre-intervention through to three months post-intervention. A validated Python-based pipeline implemented Mobilise-D gait sequence detection, wavelet-based heel strike and toe-off identification, and extraction of linear spatiotemporal parameters alongside a 24-feature nonlinear biomarker panel, which included sample entropy, multi-scale entropy variants, Lyapunov exponents, and recurrence quantification analysis metrics. A standardized clinical battery (FMA, BBS, FGA, 10 MWT, 6 MWT) was administered at each session. All three participants exhibited a 52–62% increase in Wolf’s Lyapunov exponent during SBC Week 2, which was most consistent across participants. Participant-specific entropy responses were heterogeneous, with SampEn and RCMSE showing greater longitudinal sensitivity than conventional spatiotemporal metrics. These results demonstrate that a single lumbar IMU, paired with a nonlinear analytics pipeline, can detect gait reorganization during intensive stroke rehabilitation imperceptible to standard clinical measures, providing a reproducible methodological foundation for larger controlled trials.