Grey-level morphology combined with an artificial neural networks approach for multimodal segmentation of the Hippocampus

Abstract

This paper presents an algorithm that continues segmentation from a semi automatic artificial neural network (ANN) segmentation of the Hippocampus of registered T1-weighted and T2-weighted MRI data. Due to the morphological complexity of the Hippocampus and difficulty of separating from adjacent structures, reproducible segmentation using MR imaging is complicated. The human intervention in the ANN approach, consists of selecting a bounding-box. Grey-level dilated and grey-level eroded versions of the T1-weighted and T2-weighted data are used to minimise leaking from Hippocampus to surrounding tissue combined with possible foreground tissue. The segmentation algorithm uses a histogram-based method to find accurate threshold values. Grey-level morphology is a powerful tool to break stronger connections between the Hippocampus and surrounding regions than is otherwise possible. The method is 3D in the sense that all grey-level morphology operations use a 3 /spl times/ 3 /spl times/ 3 structure element and the herein described algorithms are applied in the three directions, sagittal, axial, and coronal, and the result are then combined together.