University of Chicago

Radiology departments struggle with timely reporting for time-sensitive findings such as pulmonary embolism. Without AI triage, radiologists process exams in queue order, potentially delaying urgent cases and slowing treatment. Prior literature showed inconsistent findings on whether AI triage devices actually deliver time savings in clinical settings.

UChicago deployed Aidoc's BriefCase AI triage software, which analyzes CT pulmonary angiography (CTPA) scans to identify suspected pulmonary embolism and elevates those exams to the top of the radiologist's reading queue. The device achieved 90.6% sensitivity and 89.9% specificity. Radiologists processed reports through Microsoft (formerly Nuance) software alongside the triage tool. The study analyzed over 11,000 adult CTPA scans logged between 2018 and 2022.

During regular work hours, average turnaround time fell from 68.9 minutes to 46.7 minutes — a 32.2% reduction representing roughly 22.2 minutes saved per exam, deemed clinically significant. During off-hours, turnaround time dropped only from 44.8 to 42.0 minutes (~6.3%), a change not considered significant. A computational model (QuCAD) confirmed that time savings are highly dependent on workflow parameters such as staffing levels, exam volume, and disease prevalence.

• AI triage delivers measurable turnaround time improvements primarily during high-workload periods when radiologist queues are longest; off-hours benefit is minimal.
• Workflow parameters (staffing, exam interarrival rate, disease prevalence) significantly influence whether AI triage produces time savings — one-size-fits-all conclusions are not valid.
• Institutions should use computational modeling tools like QuCAD before deploying AI triage to predict whether their specific clinical conditions will yield meaningful efficiency gains.