Mining Skin Lesion Images with Spatial Data Mining Methods

We address the problem of identifying color variation in color skin lesion color. We adapt a spatial data mining method to this task and integrate with a segmentation method to identify significant color regions in an image. The resulting regions are compared with human perception via Kappa statistical test. Evaluation of the results indicate that the method approximates human judgment well and can be used as an automatic tool for mining skin lesion images. The approach is applicable to similar problems such as texture region identification and mining of other types of images.