THE EFFECTS OF MAMMOGRAPHY NORMALIZATION ON CLASSIFICATION

Digital mammograms should be preprocessed for computer-aided diagnosis. The aims of preprocessing are denoising and eliminating of artifacts. Mammograms are computing by different techniques and classifiers, after preprocessing step. The purpose of this study is the evaluation of the classification rates with first order textural features by different preprocessing steps. In the study digital mammograms are taken from MIAS database. The algorithm of the preprocessing step of this study includes noise clearance by median filter and artifact noise and pectoral muscle elimination by threshold techniques and morphological operations. Denoised images are normalized by a size of 512x256 pixels. Then, contrast-limited adaptive histogram equalization (CLAHE) is applied. In proposed study, mammogram images divided into 4 groups. Group 1: Original MIAS mammograms, without any preprocessing operation Group 2: Mammograms that are cleaned noises and pectoral muscles Group 3: The dimension normalized mammograms at 512x256 pixels Group 4: CLAHE applied mammograms For each group mammograms commonly used features are extracted and SVM classifier are used. According to classification results, the best classification rate is implemented by noise and pectoral muscle are eliminated groups (Group 2)