Machine Learning Based Diagnosis of Breast cancer using Mammography Data

Abstract

Tumor is an abnormal tissue which can be presented at any part of the body. It can be classified to either benign or malignant. One of the most common women's tumors infects the breast. Such tumor can be detected by mammogram which can be used by radiologist to identify a breast tumor and classify it. This is a time-consuming process and prone to error due to the complexity of the tumor. In this study, machine learning based techniques have been developed to assist the radiologist to read mammogram data and classify the tumor in a very reasonable time interval. For this purpose, several features were extracted from region of interest in the mammogram which has been manually annotated by the radiologist. These features were incorporated to a classification engine for training and building the proposed structure classification models. Dataset previously unseen to the model was used to evaluate the accuracy of the proposed system following the standard model evaluation schemes. In this thesis, three experiments have been done using five classification techniques which are Decision Tree, K-Nearest Neighbor, Support Vector Machine, Naive Bayes, and Discriminant Analysis in order to improve the performance of the model. Each experiment included several tests. Accordingly, this thesis found that there are various factors that could affect the performance, that were avoided after experimenting all the possible ways. This thesis finally recommends the use of the optimized Support Vector Machine or Naive Bayes that produced 100% accuracy after integrating feature selection and hyperparameter optimization schemes.