Residual Learning and Deep Learning Models for Image Denoising in Medical Applications

The utilization of CT scans in medical diagnostics has seen a consistent and substantial rise. However, this increased usage has raised concerns regarding the potential harmful effects of radiation exposure on patients. Reducing the radiation dose can result in more noise in the captured images, which can negatively impact the radiologist's ability to make accurate judgments with confidence. The most commonly encountered types of noise in medical images include Gaussian noise, speckle noise, and salt and pepper noise. Numerous significant efforts have been made to enhance image quality by eliminating this noise, and deep learning-based methods have gained popularity due to their effectiveness in handling various types of noise and image datasets. Within the research community, various neural network variations, such as autoencoders, generative adversarial networks (GANs), residual networks, convolutional neural networks (CNNs), and regularized neural networks, have gained immense popularity. In this paper, we comprehensively discuss eleven highly impactful approaches for image denoising based on deep learning techniques. We assess the performance of these methods using two quantitative and effective metrics: structural SIMilarity (SSIM) and peak signal-to-noise ratio (PSNR).