Abstract Background Computed Tomography (CT) is a technology that utilizes X-rays and offers several benefits in the medical field, but it comes with the drawback of radiation exposure. Radiation exposure poses risks, and extensive research is being carried out on CT to mitigate this concern. Artificial intelligence (AI)-based approaches have been introduced recently, including the application of AI-driven super-resolution (SR) techniques. However, these methods are not widely adopted, and the standards for dose reduction remain insufficient. Hence, this study aimed to assess the effectiveness of AI-based SR techniques for dose reduction. Results The experiment utilized the Lung man phantom, acquiring images by reducing the CT dose index volume (CTDI vol) by 25% from a baseline of 3.0 mGy. These images were then processed using AI-based SR models, super-resolution convolutional neural network (SRCNN) and very deep super-resolution (VDSR), to obtain the results. After learning, the results from 50% CTDI vol were generally superior to those from 25% CTDI vol . However, there was no significant difference between the outcomes from 50% CTDI vol and 75% CTDI vol . The best outcomes in this experiment were achieved by SRCNN, with a peak signal-to-noise ratio (PSNR) of 33.52 and structural similarity index measure (SSIM) of 0.878, and by VDSR, with a PSNR of 35.532 and an SSIM of 0.894. Conclusions This study aimed to evaluate the impact of dose reduction and presented AI-based SR results that vary with reduced doses. Conducted from the standpoint of dose change, the study is expected to serve as foundational data for future research in this area.
Kim et al. (Sat,) studied this question.