Chapter

Intelligent Science and Intelligent Data Engineering

Volume 7202 of the series Lecture Notes in Computer Science pp 318-325

Local Gaussian Distribution Fitting Based FCM Algorithm for Brain MR Image Segmentation

  • Zexuan JiAffiliated withSchool of Computer Science and Technology, Nanjing University of Science and TechnologyBiomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of Sydney
  • , Yong XiaAffiliated withBiomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of Sydney
  • , Quansen SunAffiliated withSchool of Computer Science and Technology, Nanjing University of Science and Technology
  • , Deshen XiaAffiliated withSchool of Computer Science and Technology, Nanjing University of Science and Technology
  • , David Dagan FengAffiliated withBiomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of SydneyCentre for Multimedia Signal Processing (CMSP), Department of Electronic, & Information Engineering, Hong Kong Polytechnic UniversityMed-X Research Institute, Shanghai JiaoTong University

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Automated segmentation of brain MR images into gray matter, white matter and cerebrospinal fluid (CSF) has been extensively studied with many algorithms being proposed. However, most of those algorithms suffer from limited accuracy, due to the presence of intrinsic noise, low contrast and intensity inhomogeneity (INU) in MR images. In this paper, we propose the local Gaussian distribution fitting based fuzzy c-means (LGDFFCM) algorithm for automated and accurate brain MR image segmentation. In this algorithm, an energy function is defined by using the kernel function to characterize the fitting of local Gaussian distributions to the local image data within the neighborhood of each pixel. A new local scale computing method is developed to estimate the variances of local Gaussian distributions. We compared our algorithm to several state-of-the-art segmentation approaches in both synthetic and clinical data. Our results show that the proposed LGDFFCM algorithm can substantially reduce the impact of by noise, low contrast and INU, and produce satisfying segmentation of brain MR images.

Keywords

MR image segmentation local Gaussian distribution fitting adaptive local scale