Longitudinal assessment of brain tumors using a repeatable prior-based segmentation

This paper presents an automatic method for a repeatable, prior-based segmentation and classification of brain tumors in longitudinal MR scans. The method is designed to overcome the inter/intra observer variability and to provide a repeatable delineation of the tumor boundaries in a set of follow-up scans of the same patient. The method effectively incorporates manual delineation of the first scan in the time-series to segment and classify a series of follow-up scans. Experimental results on 16 datasets yield a mean surface distance error of 0.22mm and a mean volume overlap difference of 12.34% as compared to manual segmentation by an expert radiologist.