Detecting Breast Cancer Based on Normalized Features of Biomarkers and Neural Networks

Abstract

Breast cancer is the most common type of cancer affecting women worldwide. Despite the extensive use of mammography as the gold standard for cancer detection and locating, biomarkers based on blood analyzes have become a promising tool to classify, diagnose, and predict the presence of all types of cancers and malignancies, especially breast cancer. Lots of machine learning algorithms have been implemented and improved to use bio-samples in breast cancer diagnosis and detection. In this paper, a new method for breast cancer detection is proposed based on replacing the raw values of biomarkers with new normalized features based on the average healthy and cancer values. These features express normalized values that indicate ratio between sample/biomarker and the average healthy and cancerous values of the same biomarker respectively. These new features will be used as input to the feedforward neural network. Coimbra Database has been used because of its importance and reliability in many research papers. The test results show a final classification accuracy of 91.7% and a sensitivity of 92.3%. The proposed method shows the ability to give a classification accuracy of more than 85% if the neural network is trained on 50% of the data only.