Simultaneous segmentation and inhomogeneity correction in magnetic resonance images

In Magnetic Resonance Imaging (MRI), intensity inhomogeneity has been an issue affecting the quality of post processing. In this paper, we present a simultaneous segmentation and inhomogeneity correction (IC) method based on active contour algorithm. It uses a generative model which is a modified Mumford-Shah functional proposed by Chan and Vese. The piecewise constant image model in the functional is multiplied by an underlying intensity inhomogeneity field. The inhomogeneity field and piecewise constant function are jointly estimated in an iterative way including solving the associated contour evolution equation and updating corresponding parameters. The algorithm is implemented using the level set framework. Test on MRI leg data shows our method achieves more accurate segmentation and IC results than other related methods in MR images with strong intensity inhomogeneity.

I. INTRODUCTION

Magnetic Resonance Imaging (MRI) is a popular tool for exploring anatomy of human body and diagnosing diseases. In the past decades, numerous computerized image processing methods have been developed to facilitate the understanding of MR images. However, intensity inhomogeneity has been an issue affecting the quality of post processing. It could be caused by spatially varying sensitivity of radio frequency (RF) imaging coils, magnetic permeability and dielectric property of imaged objects, rendering smoothly changing bright and dark regions in the magnitude images. It becomes severer in the cases of large field of view (FOV) such as whole body imaging. Under the scenario, both anatomy and inhomogeneity field contribute to the intensity value changes across the images. Without inhomogeneity correction (IC), the intensity-driven segmentation/registration methods cannot distinguish the effects of these two components and will fail.