Radiation Dose Optimisation Through Artificial Intelligence (AI)-based Auto-Thorax Collimation

Derek Yoon-Sang Lee; Nicole Chan; Richard Knight; Christopher Hayre

doi:10.1002/jmrs.70088

Radiation Dose Optimisation Through Artificial Intelligence (AI)-based Auto-Thorax Collimation

J Med Radiat Sci. 2026 Mar 27. doi: 10.1002/jmrs.70088. Online ahead of print.

Authors

Derek Yoon-Sang Lee¹, Nicole Chan¹, Richard Knight¹, Christopher Hayre^{1

2}

Affiliations

¹ Monash Imaging, Monash Health, Casey Hospital, Berwick, Victoria, Australia.
² Monash University, Clayton, Victoria, Australia.

PMID: 41891327
DOI: 10.1002/jmrs.70088

Abstract

Introduction: Artificial intelligence (AI) is increasingly adopted in digital radiography to enhance workflow efficiency, standardisation and support for radiation dose optimisation. Auto-thorax collimation (ATC) has been introduced to automate field size selection in chest X-ray (CXR) imaging; however, its clinical performance and impact on collimation consistency in routine chest X-ray practice require further evaluation.

Methods: A single-centre retrospective study analysed 400 posteroanterior (PA) erect CXRs sourced from the Picture Archiving and Communication System (PACS) and local image archives. Collimation size was measured in both the superior-inferior and lateral dimensions. An additional, 200 erect CXRs were collected to assess the repeat rate due to collimation errors. Statistical analysis was performed using two-sample t-tests for collimation measurements and a two-proportion Z-test for repeat rates. The coefficient of variation (CV) was calculated to quantify the variability in collimation field size for each operator across both ATC and manual collimation methods.

Results: ATC demonstrated tighter inferior and left lateral collimation than manual collimation (p < 0.05), while manual collimation showed tighter superior collimation. Collimation consistency was greater with ATC, as evidenced by lower CV values across operators. Repeat rates were comparable between ATC (7%) and manual collimation (8%).

Conclusion: AI-based collimation provides collimation results similar to manual performance, with improved standardisation and reproducibility. Its consistent output suggests potential benefits in high-turnover environments, enhancing workflow efficiency whilst optimising radiation safety.

Keywords: artificial intelligence; auto‐thorax collimation; collimation; dose optimisation; radiation dose; radiography.

© 2026 The Author(s). Journal of Medical Radiation Sciences published by John Wiley & Sons Australia, Ltd on behalf of Australian Society of Medical Imaging and Radiation Therapy and New Zealand Institute of Medical Radiation Technology.