Automatic identification of postural transitions using a single inertial measurement unit and dynamic time warping: a pilot study in healthy individuals and people with Parkinson's disease.

INTRODUCTION: Postural transitions (PTs) are crucial daily movements often impaired in neurological conditions, impacting autonomy and fall risk. Wearable inertial measurement units (IMUs) enable objective assessment of PTs, but robust algorithms for automatic identification remain limited. METHODS: This pilot study used Dynamic Time Warping (DTW), a time-series alignment method that is robust to temporal variations, to automatically identify PTs in healthy subjects (HS) and subjects with Parkinson's Disease (SwPD). For this purpose, 10 participants (5 HS and 5 SwPD) performed 5 postural transition tasks (sit-to-stand, stand-to-sit, supine-to-sit, sit-to-supine, and roll) using a sternum-mounted IMU. Reference PT patterns were generated from previously collected acceleration data representing optimally executed transitions. The DTW algorithm classified each detected candidate transition by minimizing its distance from predefined reference patterns. Classification performance indexes were statistically compared between groups. RESULTS: Within this pilot dataset, the DTW algorithm correctly identified 118/118 postural-transition signals in HS and 136/147 signals in SwPD. Misclassifications in SwPD primarily affected sit-to-stand (31%) and stand-to-sit (19%) transitions. DISCUSSION: These preliminary findings support the feasibility of a DTW-based approach for postural-transition identification, although larger independent validation studies are required. Future work should prioritize real-time deployment and larger validation cohorts.