Riemannian SPD learning to represent and characterize fixational oculomotor Parkinsonian abnormalities

Juan A. Olmos, Antoine Manzanera, Fabio Martínez

Abstract: Highlights•A strategy that pools convolutional representations into Riemannian representations.•A digital biomarker to quantify Parkinson’s Disease from ocular fixation videos.•An explainability strategy that highlights regions over eye video sequences.•The approach reveals remarkable performance on an extra dataset