A Computer Aided System to Discriminate Enhanced Colon Images by Three Data Mining Methods

Abstract

Computer-aided diagnosis for colon polyps automatically determines the locations of suspicious polyps and masses in Colonoscopy and presents them to doctors, typically as a second opinion. The proposed of Computer-aided diagnosis system consists:Using histogram equalization to do the image in the feature extraction and the classification. The researched image data were collected from a community hospital in Mid-Taiwan. First we used the histogram equalization to do the image enhancement, we got six characteristic values and calculate by the gray-scale co-occurrence matrix to get feature extraction. Finally, we used Decision Tree, Logistic Regression and ENSEMBLE to undergo colonoscopy image data classification. This researched found that difference of six texture parameter between normal and polyp group is significant. The accuracy of ENSEMBLE classification is best (90.00%). It indicates the ENSEMBLE classifier based on texture is effective for classifying polyp from tissue on colon imaging. The results of this study can be help the physician to get reliable and consistent diagnostic results and improve the quality of diagnostic imaging.