Hidden Markov Model based stride segmentation on unsupervised free-living gait data in Parkinson’s disease patients

Abstract Background To objectively assess a patient’s gait, robust identification of stride borders is one the first steps in inertial sensor-based mobile gait analysis pipelines. While many different methods for segmentation have been presented literature, an out-of-lab evaluation respective algorithms on free-living still missing. Method address this issue, we present comprehensive dataset, including 146.574 semi-automatic labeled strides 28 Parkinson’s Disease patients. This dataset was used to evaluate performance new Hidden Markov Model (HMM) based approach compared available dynamic time warping (DTW) method. Results The proposed HMM achieved mean F1-score 92.1% and outperformed DTW significantly. Further revealed dependency number within walking bouts. Shorter bouts ( $$< 30$$ < 30 strides) resulted worse performance, which could be related more heterogeneous increased diversity types short In contrast, reached F1-scores than 96.2% longer $$> 50$$ > 50 strides). Furthermore, showed that HMM, trained at-lab data only, transferred context with negligible decrease performance. Conclusion generalizability promising feature, as fully training might not applications. best our knowledge, large scale dataset. Our HMM-based able complexity data, thus will help enable assessment parameters future