From Spatial Domain to Learning Based Methods: A Survey on MRI and HRCT Image Denoising Methods

Medical imaging is essential for diagnosis and treatment planning, but different kinds of noise, such as Gaussian, Poisson, salt-and-pepper, and speckle noise, frequently degrades image quality. There are many radiological modalities, with each having its advantages and limitations, like MRI, CT, ultrasound, and X-ray, that provide useful information. Proper choice of radiation dosage to ensure high-quality images while reducing exposure risk is a significant issue in medical imaging. This study considers several denoising methods and assesses how well they work on datasets from magnetic resonance imaging (MRI) and high-resolution computed tomography (HRCT). Advanced deep-learning-based techniques like the denoising convolutional neural network (DnCNN) and block-matching and 3D filtering (BM3D) are contrasted with traditional techniques like bilateral filtering, guided filtering, and non-local means (NLM). The efficiency of denoising while maintaining anatomical features is evaluated by analyzing performance indicators such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and natural image quality evaluator (NIQE). The findings indicate that when compared with the other techniques, DnCNN and BM3D show better results while maintaining high visual clarity and structural fidelity. These results demonstrate the importance of selecting the appropriate denoising algorithm for the imaging modality and noise characteristics to enhance clinical decision-making and improve diagnostic accuracy.