A novel statistical scheme for the automatic detection and tracking in time of relapsing-remitting multiple sclerosis (MS) lesions in image sequences is described. Coherent space-time regions in a four-dimensional feature space (intensity, position (x,y), and time) are extracted by unsupervised clustering using Gaussian mixture modeling. The segments in the sequence pertaining to lesions are automatically detected by context-based classification mechanisms. Lesion segmentation and tracking are performed in a unified manner and not separately, as in other works. A model adaptation stage, in which space-time regions are merged, is introduced for the improvement of lesions' delineation. Qualitative and quantitative results for a sequence of 24 images are shown. The framework's results were validated by comparison to an expert's manual delineation.