Egocentric Vision Aids Freezing of Gait Detection in Parkinson's Patients

A recent investigation published on arXiv (2608.13283) delves into leveraging egocentric vision to enhance the understanding of clinical motion in everyday life, particularly concentrating on detecting freezing of gait (FOG) in Parkinson's disease (PD). This study tackles the issue that similar inertial patterns during daily activities (ADLs) can indicate either intentional pauses, interactions with objects, or movement impairments. By providing task-specific context, egocentric vision can help clarify these situations. The research involved synchronized egocentric video, wearable IMUs, and expert-annotated FOG labels from 13 PD patients in their residences. The findings revealed that an IMU-based TCN, trained from scratch, yielded the highest event-detection performance, achieving 42.3 F1 and 83.0 AUROC, surpassing the best foundation model's 32.6 F1 and 77.2 AUROC. This indicates that while foundation models are promising, a specialized IMU-based method is more effective for FOG detection, with egocentric vision potentially offering further contextual enhancements.