Machine Learning-based Classification for Unifocal and Multifocal Breast Cancer

Abstract

The early detection and accurate diagnosis of breast cancer are critical for enhancing patient outcomes and reducing mortality rates. While experienced clinicians achieve a diagnostic accuracy of approximately $79 \%$, the application of machine learning techniques has the potential to elevate this accuracy to $91 \%$. This paper focuses on the utilization of machine learning methodologies to differentiate between multifocal (MF) tumors, which are associated with a higher risk of metastasis, and unifocal (UF) tumors. A comprehensive dataset comprising 1,000 samples representing both UF and MF breast tumors, obtained through a microwave imaging technique, was utilized in this study. Nine machine learning classifiers were studied and compared in this work. Also, the convolutional neural network (CNN) was applied to the same dataset, and its performance was investigated. All the models were evaluated based on accuracy, precision, recall, and F1 score. Boosting, stacking, and voting models outperformed the other models, achieving the higher accuracy of $99.5 \%$. Additionally, results prove that adjusting the number of epochs and learning rate in CNN can play a vital role in its discrimination degree. This study underscores the significance of advanced computational techniques in the early identification of breast cancer subtypes, ultimately contributing to improved clinical decision-making and patient management.