Thesis (M.Sc.) - Cairo University - Faculty of Computers and Artificial Intelligence - Department of Computer Science

Breast cancer is one of the most common and deadliest cancer for women worldwide. However, early detection increases the chances of survival to virtually 100%. Radiologists use ultrasound images of the breast to look for signs of tumor formation such as microcal- cifications and breast masses. We aim to detect these signs using convolutional networks, a modern machine learning model that performs image classification in a single learnable step. After testing different network architectures and training configurations, we showed that convolutional networks are able to classify breast cancer with promising results. Fur- thermore, this performance will only improve as richer data sets become available. We highly encourage research in this direction. Breast cancer classification and detection using ultrasound imaging are considered a significant step in computer-aided diagnosis systems.Over the previous decades, re- searchers have proved the opportunities to automate the initial tumor classification and detection.The shortage of popular datasets of ultrasound images of breast cancer prevents researchers to get a good performance of the classification algorithms. So, data augmen- tations are used to enlarge the dataset. However, traditional data augmentation approaches are firmly limited, especially in tasks where the images follow strict standards, as in the case of medical datasets. So, a data augmentation Generative Adversarial Network (GAN) is used beside traditional augmentation.Higher accuracies are achieved when merging both methods

Issued also as CD