Artificial intelligence in CT scan range delineation: A systematic review and meta-analysis

Objectives To thoroughly investigate the impact of artificial intelligence (AI) algorithm-assisted scan range (SR) determination techniques on scan length, anatomical coverage accuracy, and radiation dose (RD) on computed tomography (CT) scans. Methods The literature published between January 2018 and May 2025 was systematically searched using five databases: EBSCOhost, IEEE Xplore, Ovid MEDLINE®, Scopus, and PubMed. Data extraction was performed by two review authors and validated by a third reviewer. The quality of the included studies was evaluated using the CLAIM and GRADE approaches. Statistical analyses were conducted using random-effects meta-analysis with standardised mean differences, assessing heterogeneity and publication bias. The findings were summarised using meta-analysis methodologies and reported descriptively. Results Six retrospective studies were included, comparing AI-assisted and manual CT SR determination techniques in terms of scan length, anatomical coverage accuracy, and RD. These studies varied in anatomical focus—including chest, abdomen, and cardiac regions—and employed diverse AI methodologies, including deep learning and machine learning algorithms. Compared to the manual SR methods, AI improved the anatomical coverage accuracy by reducing the upper and lower errors of the SR boundaries, which contributed to a mean reduction in scan length of 19.1 mm for chest CT, 42.7 mm for abdomen CT, 25 mm for chest/abdomen/pelvis CT, and 17.5 mm for coronary CT angiography (p < 0.001). However, the accuracy of automated methods varied between boundaries, with lower accuracy at inferior anatomical boundaries when compared to superior boundaries, resulting in higher over-scanning inferiorly. Despite this, all studies reported significant RD reductions with AI-based SR determination, ranging from 5 % to 47 % (p < 0.001). Conclusion AI-automatic SRs were clinically feasible and demonstrated satisfactory performance. These findings suggest that integrating such methods into clinical workflows may improve SR accuracy, reduce over-scanning, and lower RD. However, further well-designed prospective studies are needed to confirm their effectiveness across diverse clinical settings.