Clonal selection algorithm for gaussian mixture model based segmentation of 3d brain MR images

  • Authors:

  • Tong Zhang;Yong Xia;David Dagan Feng

  • Affiliations:

  • Biomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of Sydney, Australia;Biomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of Sydney, Australia;Biomedical and Multimedia Information Technology (BMIT) Research Group, School of Information Technologies, University of Sydney, Australia,Med-X Research Institute, Shanghai JiaoTong University, ...

  • Venue:
  • IScIDE'11 Proceedings of the Second Sino-foreign-interchange conference on Intelligent Science and Intelligent Data Engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Evolutionary algorithms with global search capabilities have been successfully used to replace local search heuristics in statistical image segmentation. Among them, a novel immune-inspired evolutionary method, clonal selection algorithm (CSA) has proven itself in a variety of real applications with better performance than several other evolutionary algorithms. In this paper, we incorporate the CSA into the Gaussian mixture model (GMM) based image segmentation process, and thus propose the CSA-GMM algorithm for delineating gray matter, white matter and cerebrospinal fluid in brain MR images. In this algorithm, we assume that brain voxel values to be modeled by the GMM, whose parameters are then estimated by using the CSA. Each brain voxel is then categorized by applying the voxel value and statistical parameters to the Bayes classifier. In order to improve segmentation performance by employing the spatial information, we also construct the probabilistic brain atlas for each image, and incorporate the anatomical priors embedded in the atlas into the segmentation process. The proposed algorithm has been evaluated in simulated brain MR images and been compared to the GA-EM algorithm and the segmentation routines used in the statistical parametric mapping (SPM) package and FMRIB Software library (FSL) in 18 clinical T1-weighted brain MR images. Our results show that the proposed CSA-GMM algorithm can achieve better segmentation accuracy on average.