A New Validity Index in Overlapping Clusters for Medical Images

Detecting automatically overlapping structures is a major issue in segmentation. In addition, the assessment of the quality of the clusters produced by fuzzy segmentation algorithms is one of the challenging tasks in segmentation process. To address this issue, a wide variety of functions called validity indexes have been proposed and developed. The overlap phenomenon constitute a source of failure for most of these indexes, in this context we propose a new cluster validity index V_ECS to recognize the optimal number of clusters adapted to the fuzzy c-means algorithm; based on the entropy, on the partition coefficient functions and on two criteria: the compactness within the classes and the inter-classes separation. We conducted experiments on medical imaging data sets, including simulated brain magnetic resonance imaging (MRI) and leukemia images. The experimental results reveal that, our V_ECS validity index outperform the other existing methods and has greater ability to identify the appropriate number of segments on high overlapped data sets.