Detection of Breast Cancer by Leveraging Vision Transformer Model

The most common and deadliest cancer in women is breast cancer. It ranks first among all diseases that affect women in terms of fatality and morbidity. Every year, the number of new cases has been increasing by nearly 30%. In the previous studies, convolutional neural networks (CNNs) have emerged as the preferred technique for vision applications. Though it is excelled in capturing the local spatial structures, Vision Transformers (ViT) are more advantageous in cases where contextual learning and global dependencies are critical. Due to the self-attention mechanism, it identifies the relationships among the various parts of the image. However, in order to attain the better performance, ViT need more training data as well as pre-trained model. Therefore, the authors analyzed ViT and pre-trained ViT. They conducted the experiment on the BreakHis dataset and achieved better results of 96.73%, 97.82%, and 96.8% for batch sizes of 64, 32, and 16 respectively.