Classification of multiple sclerosis lesion evolution patterns a study based on unsupervised clustering of asynchronous time-series

Based on weekly up to monthly follow-up of MS patients over one year with T2-weighted magnetic resonance images, a new clustering scheme is proposed to automatically identify lesions sharing similar temporal behaviors. The proposed method, based on spatiotemporal mean-shift and dynamic time warping, allows to detect intra and inter-patient similarities in lesion evolution patterns, which provides a quantitative approach towards the understanding of lesion evolution heterogeneity.