Researchers at Stanford University have tested an artificial intelligence model designed to reorder the CT scan queue in emergency departments, giving priority to exams most likely to reveal clinically actionable findings. The results were published in Radiology: Artificial Intelligence.
Simulation built on four years of ED data
The team trained a machine learning model on multimodal data from nearly 314,000 emergency department visits recorded between August 2020 and August 2024. The model predicts, at the moment a CT scan is ordered, how likely the exam is to show a clinically actionable result.
To test the approach, the researchers ran a discrete-event simulation calibrated to real ED operations. On a separate set of almost 21,000 CT exams, they compared the AI-based prioritization against the standard first-in, first-out queue.
A plus for urgent cases: faster results
Compared with the conventional approach, the AI-based queue produced the following changes for actionable studies:
- Median wait time reduced by 10.75 minutes
- 90th-percentile wait time reduced by 43.36 minutes
- Share of actionable results delivered within one hour: up from 48 percent to 57percent
For nonactionable studies, the picture was mixed: median wait times fell by 5.8 minutes, while 90th-percentile wait times rose by 14.46 minutes. The model reached 80 to 87 percent of the maximum possible benefit achievable with perfect predictions, according to the study.
New potential without new hardware?
Corresponding author David Kim, MD, PhD, of Stanford's Department of Emergency Medicine, and his colleagues noted that the approach could improve workflows and care by using data already generated in routine operations.
The conclusion was that 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. The study is based on a simulation and has not been tested in real-world clinical settings. All of the authors’ conclusions refer to the scenario described above.
