Novel Seed Selection and Conceptual Region Growing Framework for Medical Image Segmentation

Abstract

The objective of the paper is to propose a novel idea to improve initial conditions of seeded region growing (SRG) algorithm. We also propose a conceptual region growing framework to contribute to its progress in medical imaging. Our scheme is based on the simple observation that nature seems random but it repeats itself. Medical images are a kind of natural images and hence they must have a tendency of behaving like fractals. Our non-parametric Polygonal Seed Selection method does not need density estimation as before and shows clear Improvement to handle over segmentation problem. Qualitative results have been demonstrated on Axial Slices of Brain using traditional SRG, K-Means and Watershed segmentation.