With the development of medical imaging technology, more and more medicalimages affect medical personnel on diagnosis. In order to more effective manage andutilize these medical images, researchers pay close attention on data clustering inmedical images. While medical images are very complicated in structure and havenumerous characteristic, and most medical images are high dimensional. So manycommon data mining methods which are effective on general database do not reallyhave a good effect. It became the focus to research clustering method for medicalimages.First in this paper, it analyzes the current domestic and research on medicalimage mining. Second based on above content, it makes use of a new kind of ROIpicked up from image to description medical images’ characteristic instead of traditioninterest. This method introduces medical knowledge into clustering and hold on to ahigh precision clustering outcome.The research work and innovations of this paper are as follows:(1)Clustering for medical images based on differential evolution. We combinedifferential evolution ideology into K-means, and introduce medical knowledge intothe algorithm processes. But this algorithm needs not K parameters and is notsensitive for the initial center and noise, but does not affect the accuracy of clusteringresults.(2)Clustering for medical images based on graph theory. In the first we convertimages into a complete graph as an established form, then accomplish image’clustering by graph clustering. |