A Robust Approach to COVID-19 CT-Scan Image Denoising with Inception Residual Attention UNET

A CT scan is crucial for evaluating the severity of a COVID-19 infection; however, exposure to radiation can increase the risk of developing cancer. To mitigate this risk, the intensity of radiation is significantly reduced, though it comes at the cost of introducing noise in the CT images. UNET has gained immense popularity in medical image denoising, but as the depth of the network increases, performance begins to degrade. In this paper, a novel UNET-based architecture named Inception Residual Attention UNET (IRAUNET) for denoising single and mixed-noise CT-scan images is proposed. The improvements in existing U-Nets are as follows: (1) IRAUNET employs residual blocks to enhance the depth of the network, preventing degradation issues. (2) The proposed network introduces an inception mechanism in the residual block, allowing multi-scale feature extraction and increasing adaptability to distinct noise patterns. (3) The proposed network incorporates an attention mechanism within skip connections to refine image feature extraction and selectively preserve crucial details. Compared to seven established models commonly used for image denoising, IRAUNET exhibits substantial advancements, leading to superior numerical and qualitative outcomes.