Constraint-based region growing with local shape fitting

We proposed a novel paradigm which combines global and local image segmentations. The local segmentation algorithm is mainly combined with the global to generate accurate results. We have mainly focused on segmentation of Caudate Nucleus from brain MRI images which forms our Region Of Interest (ROI) for this research. Caudate Nucleus is a key component of basal ganglia responsible for critical brain functions and its aberrant morphology and functions have been implicated in number of important brain disorders.; The global region growing works in 3-Dimensional (3D) as well as 2-Dimensional (2D) space to segment out boundaries of ROI which resembles closely the shape of the structure. The constraints used for 3D segmentation are intensity and derivatives of intensities within the image. For 2D segmentation we use additional constraint of angular curvature of 2D growing region contour. The local region growing deforms in very near vicinity of contour segmented by global region growing algorithm, and maintains its geometry by using centerline constraint. In addition, the local region growing also uses derivatives of intensities as a constraint in guiding its deformation. We developed a novel algorithm to estimate the centerline for most of the closed shapes. At the culmination of region growing algorithms, we get CN volume segmented in 3D MR image.; We performed 18 experiments on 9 MRI datasets, two for each dataset for right and left Caudate Nucleus. These datasets are collected at 1.5 Tesla and 4 Tesla MRI machines. We chose datasets with variable resolutions and qualities. The automatic segmentation results are compared with manual segmentation done with the help of a MRI student, and results shows high overlap between two segmentations which demonstrates the high reliability and accuracy of the algorithms.