Riemannian Covariance Descriptors for Medical Signal and Image Classification

We demonstrate that replacing Euclidean distance with the affine-invariant Riemannian metric on the SPD manifold improves diagnostic classification accuracy across two clinical domains: brain tumour classification from MRI (+5.67%, p=0.016) and myocardial infarction detection from 12-lead ECG (+8.15%, p=0.008). No additional data, training, or hardware required — only a change in the distance metric. Patent pending (AU 2026902296)