Evaluate generalization of AI-assisted finding detection under routine clinical workloads

Determine whether the ability of NV-Reason-CT to identify findings initially missed by radiologists differs under routine clinical workloads involving many daily CT examinations.

Background

The preliminary reader study involved two experienced radiologists reviewing only ten CT cases, with unaided interpretation preceding AI-assisted review. Radiologists reported that they did not miss findings during the manual review, resulting in a relatively low rating for the model’s ability to identify initially missed findings. The paper explicitly leaves unresolved whether this result would change in ordinary clinical practice, where radiologists interpret substantially larger numbers of cases and may experience workload-related omissions.