Medical Image Segmentation Using Improved Affinity Propagation

Affinity Propagation (AP) is an effective clustering method with a number of advantages over the commonly used k-means clustering. For example, it does not need to specify the number of clusters in advance, and can handle clusters with general topology, which makes it uniquely suitable for medical image segmentation as most of the objects in medical images are not roundly shaped. One factor hampering its applications is its relatively slow speed, especially for large-size images. To overcome this difficulty, we propose in this paper an Improved Affinity Propagation (IMAP) method with several improved features. Particularly, our IMAP method can adaptively select the key parameter p in AP according to the medical image gray histogram, and thus can greatly speed up convergence. Experimental results suggest that IMAP has a higher image entropy, lower class square error contrast, and shorter runtime than the AP algorithm.