Wearable-Based Artificial Intelligence for Freezing ofGait Detection in Parkinson’s Disease

Freezing of Gait (FoG) is one of the most disabling motor symptoms of Parkinson’s disease (PD), characterized by brief episodes in which patients are unable to initiate or continue walking. Its unpredictable nature and strong variability across individuals make automated detection a challenging yet crucial task for clinical monitoring and personalized treatment. This thesis aims to develop robust and generalizable AI-based models for the detection of FoG using wearable inertial sensors. The research follows a unified trajectory: first, it analyzes the current landscape of IMU-based FoG datasets to identify structural limitations and establish methodological guidelines; second, it investigates activity recognition of key motor tasks—such as turning and sit-to-stand transitions—that often precede FoG episodes, providing a framework for contextualized detection. Building on these foundations, the work explores transfer learning to overcome data scarcity and reduce costs in terms of time and work, and introduces a Mixture-of-Experts strategy to enhance adaptability across heterogeneous sensor configurations and patient populations. Overall, this study contributes to the advancement of scalable and clinically meaningful approaches for Freezing of Gait detection in Parkinson’s disease, paving the way toward personalized, wearable-based monitoring systems.