Video-based Clinical Gait Analysis in Parkinson's Disease: A Novel Approach Using Frontal Plane Videos and Machine Learning.

Gait can be significantly impaired by neurological conditions such as Parkinson's disease (PD). Gait impairments can be quantified by using instrumented gait analysis techniques, but these instrumented techniques are expensive and suffer from limited accessibility in clinical practice. This paper introduces a method utilizing machine learning algorithms for video-based clinical gait analysis of frontal plane videos recorded with readily available devices. Our approach leverages open-source body pose (Google's MediaPipe) and metric depth (ZoeDepth) estimators to calculate spatiotemporal parameters (STP) of gait from videos and demonstrated high agreement and correlation with traditional marker-based motion capture (MoCap). We applied our proposed methodology to gait videos of healthy controls and people with PD (PWP). Our results agree with those obtained with high-end instrumental gait analysis techniques and support the detection of PD from gait videos.Clinical Relevance- This study presents an innovative, user-friendly, and cost-effective solution for clinical gait analysis. It will enable automated gait analysis via videos in gait-impaired individuals such as PWP, paving the way for telemedicine evaluation for movement decline. The future impact is early detection, effective remote monitoring, and new treatments for increased fall risk in an already mobility-challenged population.