A Multi-objective Framework for Brain MRI Threshold Segmentation

Abstract

In this paper, a novel framework for brain MRI threshold segmentation based on multi-objective model is proposed. Two classical techniques named Otsu's method (OTSU) and maximum entropy method (MET) are selected as the objective function based on their opposite characteristics when processing brain MRI with different levels of noise and bias field. The proposed method aims at finding trade-off solutions when segmenting images with noise and bias field. MOEA/D which has low computational complexity and high accuracy is used as the fundamental optimization tool. The Pareto front is approximated by optimizing OTSU and MET simultaneously. We employee the angle based method to find knee point as the final solution which contains more information from Pareto front. The experiments are carried on BrainWeb dataset to verify the performance of proposed framework. The segmentation results also indicate the effectiveness of the new approach.