Breast Cancer Histopathological Image Classification Using an Ensemble of Deep Convolutional Neural Networks

Women’s mortality rate from breast cancer is currently greater than other types of cancer, and the rate is rising day by day. As a result, it is vital to analyze breast cancer at an early stage to avoid tragic deaths. Traditional approaches take a long time and require the experience of expert radiologists, who may or may not agree with their results. For all of these reasons to develop an automated breast cancer recognition system, in this study we have introduced an approach using breast cancer histopathological images and a variant of AlexNet, VGG-16, ResNet-50, and Inception-v1 models and we have proposed an ensemble classifier, that is based on a majority voting mechanism. The performance of all four variants of the CNN models was evaluated using the BreaKHis dataset and the 5-fold cross-validation process. This dataset consists of images with 40×, 100×, 200×, and 400× magnifying factors. We have performed binary classification to classify benign and malignant tumors on each magnification of the dataset separately. After evaluating the performance of all four CNN models, we have selected the three best-performed CNN models based on their accuracy, precision, recall, and f1-score. Then, these three best-performed CNN models, AlexNet, ResNet-50, and Inception-v1, were used for ensemble classification using a majority-based (hard voting) voting mechanism to classify benign and malignant tumors. Our proposed voting-based ensemble classifier shows better performance with 99.87%, 99.85%, 99.98%, and 99.86% accuracy respectively for 40×, 100×, 200×, and 400× magnification of the dataset in contrast to the latest up-to-date methods for classifying tumors, both benign and malignant.