Unsupervised MR Brain Images Segmentation Using Fuzzy C-Means Based on Fuzzy Sufficient Dimension Reduction

• Activity Number: 178
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Han-Ming Wu*+
• Companies: Tamkang University
• Keywords: Fuzzy c-means clustering ; Sliced inverse regression ; Sliced average variance estimation ; Unsupervised image segmentation ; Dimension reduction

Abstract

In this study, we propose an iterative segmentation approach for MR brain images using fuzzy c-means (FCM) based on the sliced-based fuzzy sufficient dimension reduction (SDR) techniques. Firstly, FCM is conducted to obtain the initial cluster memberships of the image pixels. Then the sliced-based sufficient dimension reduction such as sliced inverse regression (SIR) and sliced average variance estimation (SAVE) are performed to extract the features of the image based on these memberships. The resulting features are served as the input for the next run of FCM. The iteration stops until the improvement is within a tolerance. The proposed method is evaluated on a set of benchmarks of the simulated and clinical MR brain images. The evaluation results indicate that this approach provides an improvement over the methods to be compared using FCM without SDR.