Abstract
Breast cancer is one of the most common cancers affecting women, characterized by the presence of malignant cells in breast tissue. Over time, these cancer cells can rapidly proliferate, spreading throughout the breast and metastasizing to other organs in the body. In recent years, the incidence of breast cancer among women has been increasing at an alarming rate. This study aims to develop methods for detecting and localizing breast lesions through the analysis of scientific research and available data. The research involves classifying lesions caused by breast cancer and differentiating them from those caused by other diseases. Several deep learning models, such as Unet, ResNet50, InceptionV3Small, MobileNetV3, Vision Transformer (ViT), and DeeplabV3+, were explored using datasets of CT images and breast cancer cell images. The study focuses on the automated classification and localization of lesions, proposes improvements to these models, and evaluates their performance. Various preprocessing techniques, such as flipping and rotating images, were employed to augment the dataset and improve model accuracy. The models were trained and tested using appropriate environments and libraries. Experimental results, including the accuracy, loss, and training time of models like DenseNet201, Unet, ResNet50, InceptionV3, MobileNetV3Small, ViT, and DeeplabV3+, were provided. The models' performances were thoroughly compared and evaluated, offering insights into their effectiveness in detecting and diagnosing breast cancer through CT imaging.