Thursday, December 4 | 9:40 a.m.-9:50 a.m. | R3-SSIN07-2 | Room E450B
In an example of a live clinical setting application, generative AI increased reporting efficiency for CT studies by nearly one-fourth, researchers have found.
Their study evaluated a tertiary care health system's experience with integrating generative AI into the radiologists' CT reporting clinical workflow.
Presenter Jonathan Huang, PhD, from the Medical Scientist Training Program at Northwestern University, will discuss the generative AI model's impact on reporting time, among other metrics. A total of 52,230 CT studies were included: 21,591 (41.3%) brain, 18,521 (35.5%) abdomen/pelvis, 11,208 (21.5%) chest, and 910 (1.7%) other studies.
The generative AI model created reports that were automatically populated within reporting software as a selectable template. Radiologists could elect to document by verifying and editing model-generated reports within their normal workflow, according to Huang and colleagues.
For their study, Huang and colleagues identified model-assisted CT reads of all anatomy from January 5, 2025, to April 15, 2025. The group also randomly sampled reads without model assistance, matched by a radiologist, just prior to the model-assisted period.
Ahead of RSNA 2025, Huang and colleagues reported a significant association of model assistance with decreased reporting time (F = 7.4; p = 0.01), estimating a per-report time savings of 5.7 minutes. Huang's group also reported no difference in the addendum rate for studies read with and without model assistance.
"This study demonstrates potential for improved care delivery with radiologist-generative AI collaboration in CT reporting," they said.
Join this session for more details from Huang's report.
