Hybrid Denoising Techniques: Noise2Void and CycleGAN for Enhanced Low-Dose CT Imaging

Medical image denoising is a critical step in improving image quality, which is essential for accurate diagnosis and treatment. However, traditional methods, including LDCT, CNN-MSE, and GAN approaches, are limited by the scarcity of large, high-quality datasets for training and the complex nature of medical images with varying noise types. To address these challenges, this research proposes a novel hybrid approach that combines the strengths of Noise2Void (N2V) and CycleGAN for unsupervised denoising of low-dose CT images. By leveraging N2V's ability to learn from noisy data and CycleGAN's capacity for image-to-image translation without paired examples, the hybrid algorithm enhances image quality while preserving diagnostic information. The proposed framework also aims to improve upon conventional methods, such as CycleGAN and GAN, by offering superior denoising efficacy and maintaining crucial structural details. Experimental results on low-dose CT scans demonstrate the effectiveness of this hybrid approach in reducing noise artifacts while preserving essential image features. The proposed N2V-based CycleGAN achieves a PSNR of 28.75, SSIM of 0.8726, and RMSE of 17.11, ensuring the highest image fidelity and clarity. The findings highlight the potential of N2V-CycleGAN models to significantly advance diagnostic imaging quality in healthcare applications, enabling safer and more accurate diagnoses.