Transfer Learning and Data Augmentation for Improved Breast Cancer Histopathological Images Classifier

Abstract

Nowadays, Breast cancer is a massive health problem worldwide. To fight against this disease, we propose a high-performance Computer-Aided Diagnosis system using deep learning. Specifically, we focus on the classification of histopathological images of breast cancer into two classes (benign and malignant). For that, we present a Mobilenet-based breast cancer classification model. This model is trained with a new extended Breakhis dataset, which is created by applying some data augmentation techniques. According to the experiments, our proposed model gives a very competitive result and the accuracy reaches 0.9. This proposed model outperforms two others proposed models based on Inception and Inception-Resnet.