Performance Analysis of Breast Cancer Classification from Mammogram Images Using Vision Transformer
Naiwrita Borah, Pooja Varma, Ashis Datta, Amish Kumar, Udayan Baruah, Palash Ghosal
Abstract
Accurate identification and diagnosis of Breast Cancer (BC) is one of the most critical and laborious challenges in the field of medical image processing. Mammography is presently the imaging modality of choice for the detection and early identification of BC in the vast majority of instances. Mammogram mass classifications (malignant or benign) remain a significant problem for radiologists and are essential to their ability to make correct diagnoses. In order to keep the tumour in check, early identification and diagnosis of BC are the most convenient and efficient ways to prevent it from becoming malignant. Manual segmentation is prone to errors due to the fact that it might be performed differently by different people. Vision transformers (ViT) have shown enormous promise in a number of computer vision applications because of their extraordinary ability to replicate long-range dependence via the self-attention process. ViT-based automation has been proposed to tackle this problem. Numerous experiments on the widely used benchmark dataset for classifying BC in mammogram images (INbreast database) confirmed the effectiveness of this architecture in comparison to current approaches, achieving a 96.48% accuracy rate with real-time performance and taking very little training time, an important condition for medical image analysis. A Graphical User Interface (GUI) is constructed for the suggested model, which can aid doctors and radiologists in making better decisions and detecting BC more quickly.