Supervised Machine Learning in Cancer Diagnosis

Abstract

Mathematics is the study of natural patterns which offers ways and tools to describe these patterns. In other words, mathematics tries to find a set of rules which mimics the behavior of a repeated or regular way in which something happens or done. In this dissertation we are going to research the patterns, in which breast cancer function, focus on Breast Cancer as it’s the most lethal type of cancer in Egypt which takes the lives of 10,000 female each year according to the World Health Organization. i Since Invasive Ductal Carcinoma (IDC) is the most common type of breast cancerii , this dissertation chooses the IDC to be its use-case in studying patterns. We focus in recognizing the IDC mathematically without human-intervention, which can automate the process of diagnosing IDC breast cancer. We used the latest technologies of Deep Neural Networks (DNN) and Convolutional Neural Networks (CNN) as well

as classical machine learning methods as K-Nearest-Neighbor (KNN), Support-Vector- Machine (SVM), Random Forest and many others and we eventually obtained an

accuracy of 85% using 90K data samples; this data were so complex and this is so close of the current state-of-the-art implementation on this data which is 87%. While doing so, we will watch our own behavior while researching a certain pattern, and tries to conclude a pattern from it, which ultimately can produces a product or an application to help doctors who are researching cancer to use it and try to find patterns mathematically even without them having programming or mathematical background.