Transferable models for clinical gait assessment
Developing gait-analysis models is limited by the amount of clinical movement data available. Collecting recordings from vulnerable patient groups is challenging, and models need to work across datasets, people and clinical tasks. This University of Toronto fellowship at the KITE Research Institute, University Health Network, and the Vector Institute explored transferable movement representations and foundation-model approaches to address that need.
CARE-PD harmonized nine Parkinsonian gait datasets from eight centres to support common benchmarks and model pretraining. A clinical application examined whether vision-based assessment could track medication-related changes in daily-life gait among care-home residents with dementia. Related collaborations produced GAITGen for synthetic Parkinsonian gait and an IFCN handbook contribution on home-based gait monitoring.