Classification of breast cancer from histopathology images using an ensemble of deep multiscale networks

Manual delineation of tumours in breast histopathology images is generally time-consuming and laborious. Computer-aided detection systems can assist pathologists by detecting abnormalities faster and more efficiently. Convolutional Neural Networks (CNN) and transfer learning have shown good results in breast cancer classification. Most of the existing research works employed State-of-the-art pre-trained architectures for clas-sification. But the performance of these methods needs to be improved in the context of effective feature learning and refinement. In this work, we propose an ensemble of two CNN architectures integrated with Channel and Spatial attention. Features from the histopathology images are extracted parallelly by two powerful custom deep architectures namely, CSAResnet and DAMCNN. Finally, ensemble learning is employed for further per-formance improvement. The proposed framework was able to achieve a classification accu-racy of 99.55% on the BreakHis dataset. (c) 2022 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.

Automated summary

Manual delineation of tumours in breast histopathology images is time-consuming and laborious, and computer-aided detection systems can improve efficiency. This study proposes an ensemble of two CNN architectures integrated with Channel and Spatial attention, achieving high accuracy in breast cancer classification.