DESNTC: Transformer-Based Double Shift Network for Breast Histopathological Image Classification
Jiqiao Liu, Xianglong Wang, Shu Zhang, Junyu Dong · 2023
Multiclass classification of breast cancer histopathology images based on deep learning methods has significant clinical implications. It can not only improve the diagnostic efficiency of pathologists but also greatly reduce the diagnostic subjectivity of pathologists. Most of the existing studies use convolutional neural networks for binary classification of benign and malignant lesions, while 8 cancer types classification is less studied. However, the study of multiclass classification is useful for clinicians to accurately distinguish different subtypes of histopathology images for accurate treatment. To this end, we propose a network architecture namely DESNTC in this work based on a transformer structure that fuses shift and convolution operations. The proposed network fully combines the ability of the transformer to extract global information with the ability of the convolution operation to mine local information and enhances the capture of channel information with the shift operation. This is the great application of the transformer model to the task of multiclass classification of breast cancer histopathological images. In this paper, studies for 8 cancer types are conducted on the BreakHis dataset. The proposed model achieves the best results among the existing studies, demonstrating the effectiveness of the approach for smart assistant diagnosis of breast cancer histopathological images.