Data Augmentation: Synthetic Image Generation for Medical Images Using Vector Quantized Variational Autoencoders

Artificial intelligence (AI) has revolutionized medical imaging, significantly improving diagnostic accuracy via the evaluation of X-rays, MRIs, and CT scans. However, the effectiveness of AI models is hindered due to high costs, privacy concerns, and challenges in acquiring large annotated datasets. Artificial data generation techniques, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and hybrid models like Disc-VAE, have gained prominence in dealing with these barriers. These techniques aim to augment datasets while preserving the complicated capabilities inherent to medical images. This paper compares the efficacy of Vector Quantized VAE with GAN and Disc-VAE using metrics Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR). The Vector Quantized VAE outperforms GAN and Disc-VAE in terms of SSIM and PSNR metrics. The findings spotlight the transformative capability of synthetic information augmentation for the medical system by enabling early and accurate disease diagnosis. These improved SSIM and PSNR metrics suggest that the generated images preserve essential diagnostic details, thereby enhancing the potential for early and accurate clinical diagnosis. It guarantees advanced performance and robustness of AI models, facilitating timely intervention and improved patient outcomes.