Transforming CT imaging with deep learning: Noise reduction, artifact management, and clinical applications – A comprehensive review

Le résumé fourni par la source

Deep Learning Reconstruction (DLR) is emerging as a significant advancement in Computed Tomography (CT) imaging by addressing crucial issues such as noise reduction, artifact suppression, and dose optimization. Traditional CT reconstruction methods, namely Filtered Back Projection (FBP) and the relatively newer Iterative Reconstruction (IR), have limitations in preserving image quality as radiation doses are lowered. While FBP allows for rapid image generation, it is highly prone to image noise, often requiring higher radiation doses to maintain diagnostic quality. In contrast, IR employs iterative refinement techniques to reduce noise and enhance image quality, but this can lead to unnatural textures that undermine diagnostic confidence especially at lower doses. DLR has shown potential for clinical applications across various imaging subspecialties, such as neuro, thoracic, abdominopelvic, cardiovascular, and pediatric imaging. DLR improves lesion detection, increases soft-tissue Contrast-to-noise, and supports lower-dose protocols, which are particularly important for pediatric patients. Compared to IR, DLR retains more fine anatomical details while effectively reducing artifacts such as beam hardening and motion distortions. However, DLR also faces challenges related to model interpretability, dataset diversity, and computational resource requirements. Addressing these issues through adaptive learning models, explainable AI frameworks, and cross-institutional data sharing will be essential for broader adoption. As DLR continues to advance, its integration with AI-driven diagnostic tools and multi-modality imaging may contribute to advancing the future of CT imaging, enhancing both diagnostic accuracy and patient safety.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.