Abstract
In this study, we present a novel hybrid model combining the Vision Transformer (ViT) and Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) for thyroid nodule detection in ultrasound images. While traditional methods, such as Convolutional Neural Networks (CNNs), have demonstrated success in image classification, they primarily focus on local features and may struggle with the global context, which is critical in medical imaging tasks. Recent studies have addressed data augmentation issues through Generative Adversarial Networks (GANs), but challenges remain with class imbalance in medical datasets. This research fills the gap by integrating ViT, which captures both local and global contextual information, with WGAN-GP, which generates high-quality synthetic images for augmenting imbalanced datasets. Our study highlights the potential of this hybrid approach to improve classification accuracy and robustness compared to existing methods. To validate the performance of the projected ViT-WGAN-GP model, a complete experimental analysis is conducted on TN5000 and UD-TN ultrasound image datasets. The comparative study highlighted the improvement of the ViT-WGAN-GP model, achieving maximum accuracies of 96.8% and 97.1% on the TN5000 and UD-TN datasets, respectively. The experimental results highlight the potential of integrating ViT-WGAN-GP for automated, reliable thyroid nodule classification, providing a promising mechanism for medical professionals in diagnostic radiology.