Image Mining using Association Rule for Medical Image dataset

The concept of data mining for discovering frequent image patterns in mammogram images using association rule is presented. Proposed method works in two phases. First phase is segmentation of digital mammogram to find region of interest (ROI). It consists of median filtering for noise removal, morphological processing for removing the background and suppressing artifacts, image enhancement techniques to improve image quality followed by region growing algorithm for complete removal of pectoral muscle. Second phase is image mining to find frequent image patterns present in mammogram images using Association rule. It consists of feature extraction, optimization by selecting most discriminating features among them, discretization of selected features and generation of transaction representation of input images. This is given as input to Apriori algorithm to generate association rules. The proposed method uses a new ESAR (Extraction of strong association rule) algorithm to obtain strong, effective and highly correlated association rules from the rules obtained using Apriori algorithm in previous step. Result shows that image mining is feasible and gives strong association rules. These association rules can be further used for effective diagnosis of mammogram images. (C) 2015 The Authors. Published by Elsevier B.V.