Classification of Breast Cancer with Breast X-Ray Images via Convolution Neural Networks, Vision Transformers and AlexNet

Abstract

Breast cancer is one of the most dangerous types of cancer, affecting many people long-term and leading to death. For this reason, it has become a frequently studied and emphasized topic in the medical field. Furthermore, significant advances in computer science have attracted the attention of the medical world, and computer science has also been incorporated into this challenging disease process to address it. The rapid development of artificial intelligence in recent years has led to a rapid increase in research on breast cancer. Numerous AI-based models have been developed to prevent human errors and to assist and support researchers in decision-making. In this study, one of these models was developed, and three different deep learning (DL) models were proposed to classify breast cancer as breast cancer-negative and breast cancer-positive. The study was adapted for computer vision (CV) using a Kaggle dataset called Breast Cancer, consisting of 3,383 breast tumor mammography images; the labels are 0 and 1, respectively, and the image dimensions are 640 x 640 pixels. In this study, three models were trained to classify breast cancer images: a Convolutional Neural Network (CNN), VisionTransformers (ViT), and AlexNet, trained with 45, 75, and 50 epochs, respectively. HuggingFace Space was used with these three models to classify breast cancer. The HuggingFace web application provided breast cancer classification based on the three models. Performance metrics, accuracy, loss, and execution time outperformed the CNN model, achieving a more optimal execution time (807.82 seconds), accuracy (0.9544), and loss (0.1078). The model has achieved significant success in breast cancer, and with further refinements, it is anticipated that the model will be suitable for use as a decision support system.