Segmentation of the hippocampus using 3D edge-based level sets with focus on initialization methods

This paper investigates the segmentation of the hippocampus using a 3D edge-based level set applied to MRI. It uses bias correction and noise removal methods tailored for MRI data. It considers different initializations of the level set algorithm to cope with the size, shape and intensity variations of the hippocampus, investigating the use multiple initializations in different sections (i.e. the head, body and tail) of the hippocampus and a `tailored' initialization constructed using superquadrics. The results of segmentation are evaluated on two public MRI datasets that include manually annotated delineations of the hippocampal boundary that we use as ground-truth to evaluate the segmentation. Experimental results demonstrate a significant improvement in segmentation performance, rising to 85% for the best combination, which exceeds the performance reported by previous authors on these datasets.