Seven commercial AI devices for lung cancer detection on chest x-rays show substantial and clinically meaningful performance variability.
Key Details
- 1Seven commercial AI tools were evaluated on 5,235 chest x-rays in a UK single-center study.
- 2Sensitivity ranged from 20.8% to 77.8%; specificity from 58.9% to 98.4%; positive predictive value from 1.5% to 28.4%.
- 3Significant differences were observed in 39 of 44 pairwise comparisons between devices.
- 4Three devices detected more tumors than radiologists; four detected fewer.
- 5False positives for tumor detection ranged widely (10 to 2,039 cases per device).
- 6Minimal agreement among devices (Fleiss κ = 0.24) highlights inconsistent results.
Why It Matters
The wide variability in performance among commercial AI products raises critical questions about device selection, safety, and standardization in clinical radiology. Establishing robust benchmarking and reporting frameworks is essential for effective and safe AI adoption in imaging.
Source
AuntMinnie
Related News
•Radiology Business
Meta Funds Ohio State’s Radiology AI Innovation Projects
Meta will fund Ohio State researchers to explore AI-driven advancements in radiology, including chest X-ray reporting and disease prediction.
•AuntMinnie
AI Triage Tool Speeds Up ICH Head CT Turnaround in ED
AI-based triage for intracranial hemorrhage on head CT reduced emergency department report turnaround by 13 minutes.
•AuntMinnie
ChatGPT-5 Performs Poorly at Detecting Distal Radius Fractures on X-rays
ChatGPT-5 missed about half of distal radius fractures on x-rays, showing limited diagnostic capability in a real-world scenario.