Accurate and Robust Active Contour Model for Medical Image Segmentation and Correction

Abstract

Magnetic resonance imaging has developed at an extremely rapid rate and magnetic resonance (MR) images can provide a lot of information for doctors to diagnose. However, MR images always suffer from intensity inhomogeneity which may cause difficulty in segmentation and analysis. This paper presents an accurate and robust active contour model for MR images. Inspired by the idea of the multiplicative intrinsic component optimization (MICO) model, we first define a data term by transforming the energy functional of the MICO model into the level set framework. Then, we define a new energy functional by incorporating a length term and an edge detector function into the data term. We present our model in both two-phase and four-phase level set formulations. The split Bregman method is applied to efficiently minimize the energy functionals. We apply our model to lots of brain MR images to test its performance. Experimental results show that our model can handle these images well even if they are polluted with serious bias field or shadows and is robust to noises. Comparison with the MICO model and the Coherent Local Intensity Clustering (CLIC) model also demonstrates the superiority of our model in terms of segmentation accuracy and correction effect.