Hierarchical Approach for Neonate Cerebellum Segmentation from MRI: An Experimental Study

Abstract

Morphometric analysis of brain structures is of high interest for premature neonates, in particular for defining predictive neurodevelopment biomarkers. This requires beforehand, the correct segmentation of structures of interest from MR images. Such segmentation is however complex, due to the resolution and properties of data. In this context, we investigate the potential of hierarchical image models, and more precisely the binary partition tree, as a way of developing efficient, interactive and user-friendly 3D segmentation methods. In particular, we experiment the relevance of texture features for defining the hierarchy of partitions constituting the final segmentation space. This is one of the first uses of binary partition trees for 3D segmentation of medical images. Experiments are carried out on 19 MR images for cerebellum segmentation purpose.