Eleven Takeaways from New Analysis of CT-Based AI for Lung Cancer Screening
In a new capabilities analysis of 16 CE-marked AI software platforms for CT lung cancer screening, researchers found that only 6.7 percent of supporting peer-reviewed studies were prospective and that none of the studies assessed impact on patient outcomes.
A new analysis of 16 CE-marked AI software modalities for CT lung cancer screening found that over half of the peer-reviewed studies for these products did not have multicenter data, a quarter of the studies assessed modalities with scans from multiple CT devices and only four studies offered prospective data.
In the study, recently published in
Here are 11 key takeaways from the analysis.
- Fourteen of the 16 reviewed AI software products provide adjunctive support for detection of solid and non-solid lung nodules on CT.
- None of the reviewed AI software platforms offer detection of endobronchial or cystic lesions.
- AI-based risk scoring was available for four of the 16 products.
- Nine of the 16 AI software modalities offered substantial task coverage (50 < 75) for nodule management recommendations but none of the products provided high task coverage (75 to 100) as per the Lung-RADS v2022 criteria.
- Six of the 16 CE-marked products had no published peer-reviewed evidence.
- Multicenter data was utilized in 24 of the 60 reviewed studies.
7. Only 25 percent of the studies had CT scan data from multiple CT device manufacturers.
8. While 70 percent of the studies assessed diagnosticaccuracy/efficacy of the AI software products, none of the studies assessed the impact of the software on patient outcomes.
9. Only four of the 60 reviewed studies had prospective research.
10. Only two of the 16 AI software products offer support for detection of juxtapleural nodules.
11. Fourteen of the 16 AI products offered measurements for solid and non-solid nodules.
Related Content
