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

The participants were only 28 people with Parkinson's disease, all walking without aids and at Hoehn and Yahr stage I-III, so it does not show how the method performs in more advanced PD or with walking aids. The reference labels came from one human annotator working directly on sensor signals, with no independent gold standard, so some labelled strides may be wrong. Only one randomly selected day from a 14-day home recording was labelled, so the free-living dataset may not represent a full two-week period. The algorithm was evaluated only after walking bouts had already been defined, so it still needs a reliable way to detect when walking starts and ends in continuous home data. Performance was lower for short bouts, and short bouts made up more than 40% of daily strides, so the overall score may hide problems in irregular walking. The study reports algorithm accuracy, not patient outcomes, fall risk, or clinical decisions.

Funding came from Deutsche Forschungsgemeinschaft, Innovative Medicines Initiative, Bayerisches Staatsministerium für Wirtschaft, Infrastruktur, Verkehr und Technologie, and Friedrich-Alexander-Universität Erlangen-Nürnberg. No author conflict-of-interest declaration is included in the supplied text.

Do not read this as evidence that wearable gait monitoring improves Parkinson's disease care. The study compared two computer methods for finding stride boundaries in sensor data, not patient outcomes, fall risk, or treatment effects.