Breast Cancer Diagnosis from Histopathological Images Using Deep Neural Networks

Breast cancer is a leading cause of death among women worldwide, highlighting the urgent need for early and accurate detection. Histopathological examination of tissue biop-sies remains the gold standard for diagnosis, but it is time-consuming and prone to human error. Deep learning models have shown promise in automating breast cancer detection from histopathological images. This study investigates the effectiveness of a Vision Transformer (ViT) model for classifying breast cancer from histology images, aiming to improve the accuracy and reliability of breast cancer detection. We exploit the BreAst Cancer Histology images (BACH) dataset comprising a large number of annotated microscopy images categorized as nor-mal, benign, in-situ carcinoma, and invasive carcinoma. While previous studies with CNNs achieved good results, we explore ViT's potential, known for capturing long-range dependencies in images. Our method involves ViT for patch-wise classification followed by an aggregation strategy to obtain image-wise labels to analyze Whole-Slide Images (WSIs). We assess the effectiveness of ViT in comparison with well-established CNN models such as VGG16 and ResNet. Our experiments demonstrate that the ViT model achieves a remarkable accuracy of 91%, outperforming its competitors. Our solution not only surpasses conventional CNN-based techniques in terms of accuracy but also lays a more solid foundation, presenting results that are more stable, reflecting superior performance not only in patch-wise analysis but also in Whole-Slide Image analysis tasks.