<![CDATA[Breast cancer remains one of the most prevalent cancers in Indonesia, and early detection plays a vital role in improving patient outcomes. Ultrasound imaging is a non-invasive and accessible technique used to classify breast conditions into normal, benign, or malignant categories. The advancement of deep learning, particularly Transfer Learning with Convolutional Neural Networks (CNNs), has significantly enhanced the performance of automated image classification. However, the effectiveness of CNNs heavily relies on large, balanced datasets—resources that are often limited and imbalanced in medical domains. To address this issue, this study explores the use of Wasserstein Generative Adversarial Networks (WGAN) for synthetic data augmentation. WGAN is capable of learning the underlying distribution of real ultrasound images and generating high-quality synthetic samples. The inclusion of the Wasserstein distance stabilizes training, with convergence observed around 2500–3000 epochs out of 5000. While synthetic data improves classifier performance, there remains a potential risk of overfitting, particularly when the synthetic images closely mirror the training data. Compared to traditional augmentation techniques such as rotation, flipping, and scaling, WGAN-generated data provides more diverse and realistic representations. Among the tested models, VGG16 achieved the highest accuracy of 83.33% after WGAN augmentation. Nonetheless, computational resource limitations posed challenges in training stability and duration. Furthermore, issues related to model generalizability, as well as ethical and patient privacy considerations in using synthetic medical data, must be addressed to ensure responsible deployment in real-world clinical settings.]]>
Copyrights © 2025
