A Kernel-based Fuzzy C-Means Algorithm with Partition Index Maximization for MRI Image Segmentation

Ozdemir and Akarum (2002) proposed a partition index maximization (PIM) algorithm and successfully applied PIM for a color quantization. Nevertheless, it still lacks enough robustness for noise and outliers, and sensitivity for fuzziness parameter m. To overcome these drawbacks, we adopt a Kernel-induced metric in the data space to replace the original Euclidean norm metric and propose a Kernel-based Fuzzy c-means algorithm with partition index maximization (KPIM). Numerical and image experiments illustrate that the proposed algorithm actually performs well. The KPIM can exhibit the robustness to outlier, noise and fuzziness parameter m. In addition, KPIM has faster computational efficiency than PIM and FCM. Finally, this algorithm is applied in segmenting MRI images. From these MRI segmentation results, we find that KPIM provides better detection of abnormal tissue than PIM. On the whole, the proposed KPIM is a robust clustering algorithm and suitable for MRI segmentation.