What does this research mean for the field?

Artificial intelligence enables clinically meaningful radiation dose reductions of roughly 40–90% across multiple computed tomography applications while preserving diagnostic confidence. Novelty: ClaimNovelty.SYNTHESIS. Consensus alignment: ConsensusAlignment.SUPPORTS_CONSENSUS.

What question did this study set out to answer?

To synthesize clinical evidence on AI-based strategies for optimizing radiation dose in CT from 2020 to 2025.

February 21, 2026

Artificial Intelligence for radiation dose reduction in computed tomography: a narrative synthesis of clinical evidence from 2020 to 2025

Key Points

To synthesize clinical evidence on AI-based strategies for optimizing radiation dose in CT from 2020 to 2025.
Systematic search of PubMed, Scopus, and IEEE Xplore
Inclusion of 86 studies based on criteria related to radiation dose and diagnostic performance
Narrative synthesis of findings grouped by anatomical region and AI application
AI-enabled substantial dose reductions of 40-90% across various CT applications
Chest imaging demonstrated 30-95% reductions supporting lung cancer screening
Paediatric applications reported 50-95% reductions, nearing radiography doses
Workflow AI features added independent benefits but showed sensitivity decline for small lesions

Abstract

Abstract Computed tomography (CT) is essential to modern clinical practice but contributes substantially to population radiation exposure, particularly in oncology, paediatric and screening pathways. In line with the ALARP principle and IR(ME)R requirements, there is growing interest in the use of artificial intelligence (AI) to reduce dose without compromising diagnostic performance. This work synthesised evidence published between January 2020 and May 2025 on AI-based strategies for CT dose optimisation, including deep learning reconstruction, denoising and workflow automation.A systematic search of PubMed, Scopus and IEEE Xplore identified 1,224 records. After removal of 239 duplicates and screening of abstracts and full texts, 86 studies met the inclusion criteria. Eligible studies reported clinical or patient-based outcomes relating to radiation dose, diagnostic accuracy, image quality or feasibility. Phantom-only work, non-CT imaging and conference abstracts without full text were excluded. Due to methodological heterogeneity, findings were synthesised narratively and grouped by anatomical region and AI application.Across indications, AI consistently enabled substantial dose reductions while maintaining diagnostic adequacy. Chest imaging demonstrated 30–95% reductions, with ultra-low-dose protocols (~0.1–0.5 mSv) supporting lung cancer screening and nodule detection. Abdominal and hepatic imaging achieved around 40–70% dose reduction with preserved lesion visibility. “Double-low” and “triple-low” vascular protocols reduced both radiation and iodine by 40–75%. Paediatric applications reported 50–95% reductions, in some cases approaching doses comparable to radiography. Workflow AI, including auto-positioning and scan-length optimisation, provided additional independent benefits.However, most studies were single-centre and vendor-supported, and sensitivity for very small or subsolid lesions declined at the lowest doses. Altered image texture at high denoising strengths and limited multicentre validation remain concerns.Overall, AI offers clinically meaningful radiation dose reductions of roughly 40–90% across multiple CT applications while preserving diagnostic confidence. The strongest evidence relates to chest screening, oncology follow-up, vascular imaging and paediatrics. Successful NHS implementation will require governance, quality assurance, training and multicentre evaluation.Keywords:artificial intelligence; computed tomography; radiation dose reduction; deep learning reconstruction; ALARP; diagnostic image quality

اسأل الذكاء الاصطناعي

Bookmark