Block-Level Cross Diagonal Texture Matrix-Based Feature Extractor for Digital Mammogram Classification

Breast cancer holds the second rank for causing high mortality and morbidity among women and still continues to be one of the major health challenges. Early detection and diagnosis are considered to be an effective way to control this death rate. Therefore, the radiologists need a computer-aided diagnosis (CAD) system as a second opinion to reach a decision for the cancer patients. The present paper deals with a CAD system which can effectively diagnose the digital mammograms as normal, benign, and malignant. Here, a block-based cross diagonal texture matrix (CDTM) is utilized for extracting the texture features from the mammograms. Further, principal component analysis (PCA) is employed to select the most relevant features which in turn are fed to the extreme learning machine (ELM) classifier to classify the mammograms. The experimental evaluations are made on two datasets, namely, MIAS and DDSM in terms of different performance measures like accuracy, sensitivity, specificity.