A comparison of different Gabor feature extraction approaches for mass classification in mammography

摘要: We investigate the performance of six different approaches for directional feature extraction mass classification problem in digital mammograms. These techniques use a bank Gabor filters to extract textural features. Directional features represent structural properties masses and normal tissues mammograms at orientations frequencies. Masses micro-calcifications are two early signs breast cancer which is major leading cause death women. For detection masses, segmentation results regions interest (ROIs) not only include but suspicious as well (which lead false positives during discrimination process). The reduce by classifying ROIs tissues. In addition, detected required be further classified malignant benign. evaluated over extracted from MIAS database. Successive Enhancement Learning based weighted Support Vector Machine (SELwSVM) used efficiently classify generated unbalanced datasets. average accuracy ranges 68 100 % obtained methods our paper. Comparisons carried out on statistical analysis make recommendations.