Classification of breast cancer molecular subtypes using convolutional neural networks

Breast cancer is one of the most dangerous diseases due to its prevalence and mortality rate. Correct classification of the disease is crucial because its heterogeneous nature. In addition to histological and morphological classification; molecular sub-type classification studies have been gaining momentum in recent years. Since the molecular subtypes affect the course of the disease, personalized treatments are being preferred. Among the molecular subtypes Triple Negative is the most aggressive and dangerous. Early diagnosis is incredibly crucial as in other cancer types. In this study DenseNET, XCeption and GoogleNet pre-trained Convolutional neural network (CNN) models have been used to classify molecular subtypes using only MRI images pixel data. Proposed method is independent of user, fast, automated and has produced significant success rate. Study aims to assist oncologists with decision support system, enable early diagnosis and reduce unnecessary biopsies. Obtained results are encouraging