A Comprehensive Review of Medical Image Denoising: Techniques, Challenges, Applications and Future Directions

The concept of noise reduction in medical images plays a vital role in improving image quality and accuracy associated with diagnosis, and facilitating reliable automated analysis. This survey presents a structured overview of medical image denoising, beginning with foundational definitions and noise models for modalities such as MRI and CT. We critically evaluate persistent gaps, including the limited generalizability of deep learning models and the lack of clinical validation frameworks while proposing future directions to bridge these gaps. Medical image denoising has evolved through three distinct eras such as traditional filtering and transform-domain methods, data-driven machine learning approaches, and modern deep learning architectures. This review presents a paradigm-centric analysis of these developments, categorizing methods by their underlying learning framework and clinical imaging modality. We highlight how Convolutional Neural Networks (CNNs) and vision transformers have surpassed classical techniques in preserving diagnostic features while reducing radiation dose. We explore commonly used benchmark datasets that support the development and evaluation of denoising algorithms emphasizing their importance in standardizing comparisons. The paper systematically evaluates the strengths of each paradigm through both quantitative metrics and clinical utility assessments. Experimental details and findings from key studies are summarized illustrating methodological effectiveness and real-world implications. The clinical applications of denoising methods are also highlighted demonstrating their utility in improving image interpretability, surgical planning, and reducing radiation dose. Despite advancements, challenges persist, including limited generalizability across imaging conditions, difficulty in acquiring clean ground truth data, and the computational demands of deep models. We also examine limitations in model transparency and the need for clinical validation. Recent trends, including self-supervised learning, hybrid models, and real-time applications, offer promising directions for future research. This review aims to guide researchers by providing a consolidated understanding of medical image denoising methodologies and their clinical significance.