A spatio-temporal model-based statistical approach to detect evolving multiple sclerosis lesions

Abstract

The effects of new treatments need to be assessed: in the case of multiple sclerosis it is possible to measure those effects by studying the evolutions of temporal lesions in time series of MRIs. But it is a laborious task to manually analyze such sets of images. This article proposes a new method to statistically analyze a series of T2-weighted MRIs of a patient with multiple sclerosis lesions taking both temporal and spatial coherence into account. The main idea of the method is to fit a temporal parametric model of intensity evolution on each voxel of the series; these estimations give different parameter values in the case of normal and pathological areas. A statistical inference stage makes it possible to determine significant sets of connected voxels corresponding to pathological evolving areas. The significancy is estimated using permutation tests. Promising results show the feasibility of our approach. On our data sets the evolving lesions were detected and their temporal behavior could be quantified.