Breast cancer classification from histopathological images using Transformers

Accurate detection and classification of breast cancer is a critical task in medical imaging due to the complexity of breast tissues. Transformers have recently impacted the computer vision area, also in the field of medical image analysis. However, breast cancer classification from histopathological images using transformers faces a big challenge: Transformers are not suitable for small sets of data and medical datasets are more difficult to obtain. Therefore, breast cancer classification from histopathological images using Transformers is of great significance. In this study, we propose a breast cancer classification method using transformers without large sets of data. The network automatically extracts features through a supervised phase from images with specified size, and presents the result as a probability matrix as either a positive sample (malignant) or a negative sample (benign). The proposed model can achieve the accuracy about 89% when training from scratch on public dataset BreaKHis. The simple and compact model is made accessible to those equipped with basic computing resources and trained them in less than half hour. Consequently, the proposed method is better than the traditional ones, as it automatically learns the best possible features and experimental results show that the model outperformed the previously proposed transformers.