Generative Deep Learning for Data Augmentation of Ultrasound Breast Cancer Scans
Radwa Taha, Rana Sabry, Mohammed A.‐M. Salem, Shereen Moataz Afifi · Journal of Engineering and Science in Medical Diagnostics and Therapy · 2025
Abstract Breast cancer is one of the leading causes of death in women worldwide; due to its high mortality rate, early detection is key in death prevention. Machine learning and deep learning have been investigated to aid in the early diagnostic process. However, due to the limited availability of breast ultrasound (U.S.) datasets, researchers cannot obtain a good performance of classification algorithms. Traditional augmentation approaches have been applied to mitigate data availability, but they are firmly limited, especially in tasks where images follow strict standards, as in the case of medical datasets. Therefore, in addition to traditional augmentation, we apply a new methodology for data augmentation using generative adversarial networks (GANs). The proposed methodology integrates two types of GAN models with traditional augmentation to achieve high results. The first is a simple GAN model, and the second is a deep convolutional generative adversarial network (DCGAN) model. This methodology was applied to three datasets. The breast ultrasound images (BUSI) dataset is from Baheya Hospital for Early Detection and Treatment of Women's Cancer in Cairo, Egypt; Dataset B is obtained from the Diagnostic Center of the Parc Tauli Corporation, Sabadell, Spain; and RODTOOK dataset was collected in the Biomedical Engineering Unit at Sirindhorn International Institute of Technology, Thammasat University, Thailand. After training, the results were evaluated both qualitatively by an expert and quantitatively using kernel maximum mean discrepancy (kMMD) that reached a value of 0.65, alongside achieving a significant classification accuracy of 95%.