When a department evaluates an AI tool, the question asked is usually whether it saves time overall. Today's data show why that is the wrong question.
The same tool, in the same institution, cut reporting time by 41% for ECG-gated thoracic CT and increased it by 7% for emergency department examinations. Averaged together those become a respectable 14.6% saving that conceals a group of readers for whom the tool is a net cost. In the emergency setting, where the clinical question is rarely the nodule, a flagged incidental finding must still be reviewed, described and dispositioned.
So when assessing any AI deployment, break the numbers down by examination type, requesting specialty and patient location before deciding where to enable it. And if a tool can be enabled selectively rather than across all studies, that is usually worth the configuration effort, because the benefit is not evenly distributed and neither is the harm.
- Measure AI benefit by examination type and requesting specialty, not overall
- The same tool sped up gated CT by 41% and slowed ED reporting by 7%
- Averages hide subgroups for whom the tool costs time
- Enable selectively where the AI's task matches the clinical question
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free