Automatic segmentation of basal ganglia iron deposits from structural MRI

Brain iron deposits have recently been suggested as biomarkers for small brain vessel diseases. Here, we present a novel, automated method for segmenting brain iron deposits in the basal ganglia from structural MRI data. It is based on minimum-variance clustering of intensities from T1and T2∗-weighted volumes, and a supervised cluster selection algorithm. This method was evaluated with MR data from 24 subjects and compared with iron deposit masks segmented manually by an experienced rater. A median Jaccard similarity index of 0.64 between manual and automatically generated segmentation masks is promising and encourages further investigations to improve the computing speed and accuracy of the method.