A comprehensive review of CT artifact reduction: From traditional reduction techniques to deep learning methods

Computed Tomography (CT) artifact reduction aims to restore image structures degraded by diverse sources, including metal implants, patient motion, and detector imperfections, and is essential for maintaining diagnostic reliability. Despite substantial progress and the rapid emergence of deep learning-based approaches, a unified and well-structured synthesis of existing techniques remains lacking. To address this gap, this review classifies CT artifacts into ring, metal, motion, and scatter types, and provides a structured overview of their underlying physical mechanisms and corresponding imaging manifestations. Following the complete CT imaging chain, from hardware acquisition and system calibration to projection- and image-domain correction, we provide an in-depth review of traditional reduction strategies and establish a structured framework for classical techniques. Subsequently, deep learning-based artifact reduction methods are grouped into supervised, semi-supervised, and unsupervised paradigms, with representative studies reviewed in terms of their methodological contributions and inherent limitations. Finally, this review discusses the clinical applications of existing methods, highlights critical challenges, and outlines several promising directions for further research.