Automated hippocampus segmentation with the Channeler Ant Model: Results on different datasets

Abstract

The hippocampus segmentation in Magnetic Resonance (MRI) scans is a relevant issue for the diagnosis of many pathologies. The present work describes a fully automated method for the hippocampal segmentation and discusses the results obtained on three datasets provided by different institutions and referring to different pathologies that involve hippocampus anatomy. The algorithm is based on an extension of the Channeler Ant Model, a powerful non linear segmentation tool belonging to the family of ant colony-based models, whose application to medical image processing already provided some promising results in the analysis of CT and PET scans. In this application, thanks to a modified pheromone deposition rule, both the grey matter intensity and the expected average hippocampus shape are taken into account. In this paper, the results on the three available datasets, obtained from the comparison to manual segmentations by different subjects and protocols, are shown: an average Dice Index in the 0.72- 0.79 range, depending on the analysed dataset, is reached.