Enhanced Deep Learning Approach for Predicting Invasive Ductal Carcinoma from Histopathology Images

Breast cancer is the more prevalent form of cancer among women, and the most common type of breast cancer is the invasive ductal carcinoma (IDC). Accurate identification and categorization of breast cancer subtypes have major importance in clinical tasks, and automated approaches are relevant in saving time and reducing error. Deep learning has been applied to several breast cancer detection tasks. It outperformed traditional approaches that include handcrafted features for data representation and machine learning methods for learning task. In this paper, we develop a deep learning architecture for the prediction of IDC. This study trained an improved CNN network and investigated the performance of the model on the IDC patch-based classification task. Experimental results show that our approach yields the best performance on the IDC dataset when compared to other published approaches. Our model achieves f-score of 85.28% and balanced accuracy of 85.41% with increase improvementof11.51% on f-score and 0.86% on balanced accuracy against the latest published deep learning approach on IDC detection.