Magnetic Resonance Image Segmentation by Contextual Fuzzy Clustering

Article type: Research Article

Authors: Kehtarnavaz, N. | Chung, M. | Hayman, L.A. | Wendt III, R.E.

Affiliations: Electrical Engineering Department, Texas A&M University, College Station, Texas 77843 | Department of Radiology, Baylor College of Medicine and Ben Taub General Hospital, Houston, Texas 77030 | Department of Radiology, Baylor College of Medicine and The Methodist Hospital, Houston, Texas 77030

Abstract: This article presents a fuzzy c-mean clustering algorithm, named contextual thresholded fuzzy c-mean, for the segmentation of magnetic resonance brain images. This algorithm incorporates contextual information into the thresholded fuzzy c-mean algorithm. This is done by adjusting the membership of voxels adaptively based upon the membership context of the surrounding voxels. The performance of this algorithm was compared both to crisp and to fuzzy clustering algorithms using “phantom” data and a set of clinical examples. The contextual fuzzy c-mean algorithm gave the best results of those tested when the criteria were accuracy of volume measurements and homogeneity of brain tissue types. A graphical user interface was developed for these methods to provide an easy-to-use software tool for clinical environments.

DOI: 10.3233/IFS-1993-1404

Journal: Journal of Intelligent and Fuzzy Systems, vol. 1, no. 4, pp. 295-305, 1993