Deep-learning models achieved comparable or superior accuracy to experienced radiologists in detecting pancreatic cancer on CT scans, especially for small tumors.
Key Details
- 1Study analyzed 2,251 patients with contrast-enhanced and noncontrast CT scans from 2007-2022.
- 2Deep learning models identified both direct and indirect CT findings of pancreatic cancer (e.g., ductal changes, atrophy).
- 3On contrast-enhanced CT, model AUC was 0.99, matching the radiologist mean.
- 4On noncontrast CT, model AUC was 0.93, outperforming the reader mean of 0.91.
- 5Sensitivity for tumors ≤20 mm reached 98% with contrast CT (model) vs. 82.6% (readers), and 86% vs. 41.1% on noncontrast CT.
- 6Models reduced analysis time to 20-22 seconds per patient vs. 63-78 seconds for human readers.
Why It Matters
This study suggests that AI tools could act as sensitive second readers, improving early detection of pancreatic cancer—a disease with extremely poor prognosis if identified late. The improved sensitivity, especially for small lesions, could shift outcomes and expedite workflow, but further prospective validation is required.
Source
AuntMinnie
Related News
•Radiology Business
Radiology AI Devices at Elevated Risk for FDA Recalls, Study Finds
Radiology AI devices are more likely to face FDA recalls, largely due to deviations from intended use and incomplete clinical data.
•AuntMinnie
Survey at SIIM Highlights Benefits and Adoption of Interactive Multimedia Reporting in Radiology
A University of Virginia survey presented at SIIM revealed radiologists have strong positive attitudes toward interactive multimedia reporting, citing efficiency and clarity gains.
•AuntMinnie
MyChart Imaging Access Links Reduce Radiologist Workload and Improve Patient Care
Embedding imaging access links in MyChart portals significantly reduces manual workload for radiologists and increases patient engagement.