DRDA-Net: Dense residual dual-shuffle attention network for breast cancer classification using histopathological images

Breast cancer is caused by the uncontrolled growth and division of cells in the breast, whereby a mass of tissue called a tumor is created. Early detection of breast cancer can save many lives. Hence, many researchers worldwide have invested considerable effort in developing robust computer-aided tools for the classification of breast cancer using histopathological images. For this purpose, in this study we designed a dual-shuffle attention guided deep learning model, called the dense residual dual-shuffle attention network (DRDA-Net). Inspired by the bottleneck unit of the ShuffleNet architecture, in our proposed model we incorporate a channel attention mechanism, which enhances the model's ability to learn the complex patterns of images. Moreover, the model's densely connected blocks address both the overfitting and the vanishing gradient problem, although the model is trained on a substantially small dataset. We have evaluated our proposed model on the publicly available BreaKHis dataset and achieved classification accuracies of 95.72%, 94.41%, 97.43% and 98.1% on four different magnification levels i.e., 40x, 1000x, 200x, 400x respectively which proves the supremacy of the proposed model. The relevant code of the proposed DRDA-Net model can be foundt at: https://github.com/SohamChatt opadhyayEE/DRDA-Net.

Automated summary

Breast cancer is caused by uncontrolled growth and division of cells in the breast, leading to the formation of a tumor. Many researchers have developed computer-aided tools for breast cancer classification using histopathological images to detect it early. In this study, a deep learning model called DRDA-Net was designed, which incorporates attention mechanism and dense connected blocks to improve classification accuracy.