A Morphological Approach for Infant Brain Segmentation in MRI Data

Abstract

This paper describes a skull stripping method for premature infant data. Skull stripping involves the extraction of brain tissue from medical brain images. Our algorithm initially addresses the reduction of the image artefacts and the generation of the binary mask that is used in the initialisation of a region growing brain segmentation process. After segmenting the brain tissue, we detail two novel post processing steps. First, we refine the edges using Kapur entropy, Low Pass Filter and gradient magnitude. Second, we remove the lacrimal glands by applying shape detection, morphological operators and Canny edge detection. The performance evaluation was conducted by comparing the segmented results with the ground truth data marked by our clinical partners.