A New and Fast Bias Correction and Segmentation Model with Application to Human Brain MR Images

Abstract

Magnetic resonance (MR) images have been used in detecting human brain disease more and more in recent years. The most important problem in the diagnose process is to find accurate regions, which makes image segmentation necessary and important. Though there are kinds of segmentation approaches for MR images, different limits and defects also appear in the practical application for bias fields and complex structures in MR images, such as inaccurate segmentation, over-correction and long iteration time. Considering those shortcomings, this paper presents a new level set model by expressing the image intensity in a continuous form and applying the split Bregman method. Thanks to the two improvements, this model can segment and correct MR images simultaneously and efficiently, and can achieve bias correction results without over-correction. Then the authors applied this model to a large quantity of MR images, and gained many good results. For better observation, quantitative and qualitative comparisons have been done between this model and the multiplicative intrinsic component optimization (MICO) model. The experimental results indicate that this model has performed well for challenging intensity inhomogeneity problems.