Unsupervised Medical Image Denoising Using CycleGAN: Improving Low-Dose CT Image Quality

Medical imaging plays a significant role in clinical diagnosis, but the use of low radiation doses in CT imaging can lead to poor image resolution and precise diagnosis. In this investigation, a novel method has been suggested to enhance low-dose CT (LDCT) images using an unsupervised deep learning model based on CycleGAN. The CycleGAN architecture enables us to perform image-to-image translation without the need for paired data, making it suitable for medical image denoising where matching high-dose CT (HDCT) images are often unavailable. Our model employs two generators for LDCT-to-HDCT and HDCT-to-LDCT translation, along with two discriminators to differentiate real and generated images in both domains. Extensive experiments demonstrate that our approach significantly improves the quality of LDCT images, yielding highfidelity HDCT-like images. This technique holds great promise for enhancing medical image quality and ultimately benefiting patient care.