Improving breast cancer classification by dimensional reduction on mammograms
Mohammad Kazem Ebrahimpour, Hamid Mirvaziri, Vahid Sattari-Naeini · Computer Methods in Biomechanics and Biomedical Engineering Imaging & Visualization · 2017
Breast cancer is known as one of the most common causes of mortality among women. Early detection increases the patient’s chances for survival while late detection may lead to death. Since cancerous and normal tissues are similar, mammogram analysis is still a challenging research. A Computer-Aided Detection/Diagnosis (CADe/CADx) system can be beneficial to enhance the quality of mammograms and increase the accuracy of early detection masses. In this paper, several dimension reduction approaches are applied on mammograms in order to enhance the prediction power in a typical CADe system. The first approach utilises Continuous Wavelet Transform for feature extraction in the first step then, the extracted features are fed into four different classifiers in the terms of power. The second method uses Fast Correlation Based Filter as a feature selection technique to select informative features. The selected features are evaluated towards different classifiers to classify mammograms. In the third approach the effect of dimension reduction is analysed. The main goal of this paper is determining the role of dimension reduction in cancer classification in the capacity of mammograms. For evaluating the proposed methods, they are compared with twenty four state-of-the-art algorithms on Digital Database for Screening Mammography. The obtained results confirm the effectiveness of the proposed methods in terms of accuracy and dimension reduction.