Radiology Partners ran its AI report drafts over 237,905 exams its radiologists had already read, and says the drafts found about 40% more across 50 conditions. This is every number the release publishes, on one chart. The average is the least informative of them.
Primary source: Radiology Partners, “Vision Language Model-Powered Radiology Reporting Shows Gains in Detection and Efficiency in Research Study of Over 237,000 Exams,” Nov. 11, 2026. Company release on an internal, retrospective, preliminary study; “Investigational device. Not for sale in the United States.” Medicare volumes: CMS Physician & Other Practitioners by Geography & Service, CY2024, via MIMI Labs.
View
80/20 · what the chart says
“About 40%” is an average over 50 conditions whose lifts run from to . The release names of the 50; the other are only inside the ranges. Every figure is relative: 57.5% more lung nodules than the original reports flagged, with no count of how many that was. A 57.5% lift on 5 nodules per 1,000 films and on 50 per 1,000 are very different Mondays for whoever orders the CT. Medicare alone billed chest X-rays in 2024. On errors, the chest X-ray drafts were error-free 91.2% of the time against 83.0% for the original reports: about fewer reports with an error per 1,000.
Read this before quoting a number
It’s a press release. Internal study, labeled preliminary, no journal paper yet. The company says the papers are on its RSNA event hub.
Found isn’t helped. The comparison is draft vs. original report, checked by blinded radiologist panels on a random subset where either one flagged a finding. No patient outcome was measured: not cancers caught earlier, not harms from extra workups.
Error measures differ by modality. Chest X-ray is “error-free” (any error); musculoskeletal and abdomen X-ray is “clinically significant error.” CT has no published figure, only “comparable to or lower.”
The draft isn’t the final report. A radiologist edits and signs it. The study scored unedited drafts against unassisted reports, not the edited product.
Investigational, not cleared. FDA authorization is still underway, per the release.