LOW DOSE CT IMAGE DENOISING USING CYCLE CONSISTENT ADVERSARIAL NETWORKS

Abstract

Low-dose computed tomography (CT) imaging is a valuable diagnostic tool but often
produces noisy images that can impede accurate interpretation. In this study, we
propose a novel approach for denoising low-dose CT images using cycle-consistent
adversarial networks (CycleGAN), a deep learning technique. By training the
CycleGAN on unpaired datasets of low-dose and high-dose CT images, our method
learns to effectively reduce noise and enhance image quality without relying on
explicitly paired training examples. We evaluate our approach on a diverse dataset of
low-dose CT scans, comparing it with traditional denoising methods. The
experimental results demonstrate that our CycleGAN-based denoising method
achieves significant noise reduction and improves the overall image quality.
Furthermore, the denoised images exhibit visual coherence and preserve important
structural details. The promising outcomes of our study indicate the potential of
CycleGAN for enhancing low-dose CT imaging, leading to improved diagnostic
accuracy and patient care. Our work contributes to the advancement of denoising
techniques in medical imaging, showcasing the applicability and effectiveness of deep
learning approaches in addressing the challenges of noisy low-dose CT images.