A New Hybrid Breast Cancer Diagnosis Model Using Deep Learning Model and ReliefF

Abstract

Breast cancer is a dangerous type of cancer usually found in women and is a significant research topic in medical science. In patients who are diagnosed and not treated early, cancer spreads to other organs, making treatment difficult. In breast cancer diagnosis, the accuracy of the pathological diagnosis is of great importance to shorten the decision-making process, minimize unnoticed cancer cells and obtain a faster diagnosis. However, the similarity of images in histopathological breast cancer image analysis is a sensitive and difficult process that requires high competence for field experts. In recent years, researchers have been seeking solutions to this process using machine learning and deep learning methods, which have contributed to significant developments in medical diagnosis and image analysis. In this study, a hybrid DCNN + ReliefF is proposed for the classification of breast cancer histopathological images, utilizing the activation properties of pre-trained deep convolutional neural network (DCNN) models, and the dimension-reduction-based ReliefF feature selective algorithm. The model is based on a fine-tuned transfer-learning technique for fully connected layers. In addition, the models were compared to the k-nearest neighbor (kNN), naive Bayes (NB), and support vector machine (SVM) machine learning approaches. The performance of each feature extractor and classifier combination was analyzed using the sensitivity, precision, F1-Score, and ROC curves. The proposed hybrid model was trained separately at different magnifications using the BreakHis dataset. The results show that the model is an efficient classification model with up to 97.8% (AUC) accuracy.