Enhance Breast Cancer Diagnosis Using Deep Learning Models On Mammogram Images in Saudi Arabia

Although 50% of breast cancer patients in Saudi Arabia are discovered at an advanced stage, early identification of the disease greatly increases therapy outcomes and survival rates. For the diagnosis of breast cancer, this work contrasts the performance of three state-of- the-art deep learning models: Vision Transformer (ViT), Siamese Recurrent CNNs (SIAM R-CNN), and CNNs. Standard measures including accuracy, precision, recall, F1 score, area under the Receiver Operating Characteristic (ROC) curve were used to evaluate performance. Furthermore, this work highlights the most important areas in mammography images for their predictions using the Explainable AI (XAI) method, ScoreCAM, therefore addressing the interpretability problems of deep learning models. While ScoreCAM visuals offer important insights into the decision-making process, hence strengthening confidence and transparency, the preliminary results reveal that the models based on Vision Transformer and ResNet-50 have greater performance in terms of precision and recall for several classes. These results show the possibility of transforming breast cancer screening procedures and enhancing early detection by use of advanced deep-learning models combined with interpretability approaches.