OTID: optimal transport-based low-dose CT image denoising

Low-dose computed tomography (LDCT) is important for reducing radiation exposure but often results in noisy images that can affect diagnostic accuracy. To address this, we propose a denoising method called OTID (Optimal Transport-based Image Denoising). OTID uses a Wasserstein GAN with gradient penalty, combining advanced attention mechanisms and a robust encoder-decoder structure to reduce noise while preserving important image details. We tested OTID on the LoDoPaB-CT dataset, comparing it to other popular methods like DnCNN and CGAN. Our model achieved the highest quality scores, with a Peak Signal-to-Noise Ratio (PSNR) of 31.32 and a Structural Similarity Index (SSIM) of 0.72, outperforming the other methods.