Impact of AI-triage on radiologist read-times

Determine whether deployment of an AI-triage device that prioritizes suspected pulmonary embolism in CT pulmonary angiography (e.g., BriefCase by Aidoc Medical) affects radiologist read-time performance, and quantify the magnitude and direction of any changes in the mean and distribution of read-times used in clinical workflow modeling.

Background

To parameterize the QuCAD queueing-based simulation, the paper needed radiologist read-time inputs. The authors assumed radiologist read-time performance is unaffected by AI-triage and therefore estimated read-times using only pre-AI data. This choice was made because the effect of AI-triage on read-times was explicitly stated as unknown. Given that read-time is a key workflow parameter influencing predicted time-savings, clarifying whether AI-triage changes read-times is important for accurate modeling and interpretation across clinical settings.

Radiologist read-times were indirectly estimated from PACS case-closure timestamps and modeled under an exponential assumption, with separate read-time distributions for PE-positive CTPA, non-PE-positive CTPA, and non-chest/CT exams. Establishing whether AI-triage modifies these read-time distributions would improve the fidelity of QuCAD predictions and generalizability of time-savings estimates.