A Robust Brain MRI Segmentation and Bias Field Correction Method Integrating Local Contextual Information into a Clustering Model

Abstract

The segmentation results of brain magnetic resonance imaging (MRI) have important guiding significance for subsequent clinical diagnosis and treatment. However, brain MRI segmentation is a complex and challenging problem due to the inevitable noise or intensity inhomogeneity. A novel robust clustering with local contextual information (RC_LCI) model was used in this study which accurately segmented brain MRI corrupted by noise and intensity inhomogeneity. For pixels in the neighborhood of the central pixel, a weighting scheme combining local contextual information was used to generate the corresponding anisotropic weight to update the current central pixel and thus remove noisy pixels. Then, a multiplicative framework consisting of the product of a real image and a bias field could effectively segment brain MRI and estimate the bias field. Further, a linear combination of basis functions was introduced to guarantee the bias field properties. Therefore, compared with state-of-the-art models, the segmentation result obtained by RC_LCI was increased by 0.195 0.125 in terms of the Jaccard similarity coefficient. Both visual experimental results and quantitative evaluation demonstrate the superiority of RC_LCI on real and synthetic images.

Magnetic Resonance Imaging Segmentation via Weighted Level Set Model Based on Local Kernel Metric and Spatial Constraint

“…Many segmentation methods have been presented by researchers to segment MRI images corrupted by intensity inhomogeneity, noise, and so on. These methods can be divided into several mainly categories: boundary-based methods [ 15 , 16 , 17 , 18 , 19 ], threshold-based methods [ 20 , 21 , 22 ], clustering methods [ 23 , 24 , 25 , 26 , 27 ], region growing methods [ 28 , 29 ], graph cuts methods [ 30 , 31 , 32 ] and deep learning method [ 33 , 34 ].…”