Statistical analysis of longitudinal MRI data: Applications for detection of disease activity in MS

Abstract

We present a method to detect intensity changes in longitudinal volumetric MRI data from patients with multiple sclerosis (MS). Preprocessing includes spatial and intensity normalization. The intrasubject intensity normalization is achieved using a polynomial least trimmed squares method to match the histograms of all images in the series. Viewing the detection of disease activity in MRI as a change-point problem, we present two statistical tests and apply them to a patient’s series of grey-level images on a voxel-by-voxel basis. Results are compared with manual lesion segmentation for one MS patient scanned approximately every 5 months for 5 years. Results are also shown for 12 MS patients with 30 monthly scans.