3-D Neuroimaging Segmentation

Abstract

We present automated anatomical segmentation of 3D neuroimaging of head Magnetic Resonance Imaging (MRI) scans. Delineation of the anatomical structures of neuroimaging data can play an important role in measuring, analyzing, and visualizing the brain's anatomical structure, changes, tumor detection, assist in surgical planning, and cognitive neuroscience research. We process MRI scans for intensity-based anatomical segmentation by performing multilevel preprocessing steps (brain extraction, bias field reduction, and partial volume effect removal). We enhance the boundary guidelines by using gradient magnitude and then reproduce the state of art 2-D efficient graph-based image segmentation technique to work on 3-D images and produce guided segmentation in O (n log n) time complexity. We test our algorithm's performance in different applications like brain anatomical segmentation, brain tumor detection, hippocampus segmentation, and MRI sequencing techniques comparison. Dice coefficient and Jaccard's coefficient are used to validate segmentation results with ground truth and individual segmentation.