Radiologists who are most experienced at reading virtual colonoscopy studies may not deliver the most accurate results, according to a pilot study presented Monday.
Radiologists who are most experienced at reading virtual colonoscopy studies may not deliver the most accurate results, according to a pilot study presented Monday.
The study compared the performance of six gastrointestinal radiologists with varying levels of expertise in diagnosing colorectal neoplasms with CT colonography. The most experienced had between 40 and 50 cases under their belts, while radiologists with moderate experience had logged 20 to 30 cases. The remaining pair of radiologists had read zero to 20 CT colon exams.
Radiologists with moderate experience demonstrated the best overall performance, showing better overall accuracy than either of the other two groups, said Dr. Martina Morrin, an assistant professor of radiology at Beth Israel Deaconess Medical Center in Boston.
The moderately experienced radiologists spent an average of 20 minutes per exam, while those with the most experience spent between six and nine minutes interpreting studies.
As the study progresses, researchers will consider the role that overconfidence, as well as on-the-job pressure to expedite results, can play in affecting diagnostic accuracy of CT colonography, Morrin said.
Moreover, the results call into question whether the established threshold of 50 cases as evidence of expertise should be reconsidered, she said. A higher threshold may be warranted in terms of both number of cases and time spent reviewing studies.
"These surprising results tell us that the amount of time spent per study, and not the radiologist's experience, has a bigger influence on diagnostic accuracy," Morrin said. "An average of 20 minutes per study tends to yield a more accurate result, while anything under 10 minutes is likely to render a poor result."
Can CT-Based Deep Learning Bolster Prognostic Assessments of Ground-Glass Nodules?
June 19th 2025Emerging research shows that a multiple time-series deep learning model assessment of CT images provides 20 percent higher sensitivity than a delta radiomic model and 56 percent higher sensitivity than a clinical model for prognostic evaluation of ground-glass nodules.