Simulation of AI-driven CT Imaging Queue Prioritization in the Emergency Department.
Authors
Affiliations (3)
Affiliations (3)
- Department of Emergency Medicine, Stanford University, 900 Welch Rd, Ste 350, Palo Alto, CA 94304.
- Department of Radiology, Stanford University, Stanford, Calif.
- Department of Biomedical Informatics, Harvard Medical School, Boston, Mass.
Abstract
Purpose To evaluate the impact of an AI-driven CT queue prioritization system on emergency department (ED) CT wait times using a discrete-event simulation (DES). Materials and Methods Multimodal data from 313,966 ED visits (August 2020-August 2024) were retrospectively analyzed. A LightGBM model was trained to predict the clinical actionability of CT studies at the time of order. DESs, calibrated to real-world operations, were conducted to compare a first-in-first-out (FIFO) policy against an AI-driven prioritization policy on the internal test set of 20,795 studies (March-August 2024). Wait times for actionable and nonactionable studies were compared between the FIFO and AI-based dispatch policies, with bootstrap 95% confidence intervals (CIs), and ceiling analyses using perfect predictions. Results Compared with the FIFO policy, the AI-based dispatch policy reduced median wait times for actionable studies in the internal test set (7,637/20,795, 36.73%) by 10.75 minutes (95% CI: -12.40, -9.10) and 90th-percentile wait times by 43.36 minutes (95% CI: -50.58, -36.80). The proportion of actionable findings obtained within one hour increased from 48.33% (3,691/7,637) to 57.30% (4,376/7,637). For nonactionable studies, the median wait time was reduced by 5.80 minutes (95% CI: -7.00, -4.40), but the 90th percentile wait time increased by 14.46 minutes (95% CI: 6.40, 23.20). The model captured 80-87% of the maximum benefit achievable with perfect predictions. Conclusion AI-driven CT imaging queue prioritization can reduce the time to diagnosis for critical findings with minimal disruption to lower-acuity patients, using existing data streams and without the need for additional hardware. ©RSNA, 2026.