- Design
- Retrospective analysis within a screening trial (UKLS)
- Population
- 361 participants with 3-month follow-up low-dose CT; 339 persisting nodules
- Primary outcome
- Automated longitudinal nodule matching success
- Effect
- 83.5% matched (79.2% to 87.1%); 1.5% of true nodules needed manual correction
This UK Lung Cancer Screening trial analysis, published 29 August in European Radiology, ran an AI nodule system fully automatically on all 361 participants who had a three-month follow-up low-dose CT, using the NELSON 2.0 threshold of a solid component of at least 100 mm³.
The AI found 378 baseline nodules in 181 participants; 39 resolved. It matched 83.5% of the 339 persisting findings across scans (95% CI 79.2% to 87.1%) — 91.8% in people with a single nodule, 72.8% in those with more than five. Of the 56 unmatched findings, 91.1% were not nodules at all, mostly pleural plaques. Only five real solid nodules (1.5%) needed manual matching.
Accurate automatic matching is what makes automated volume doubling time possible. For programmes with high throughput, this suggests AI can remove most of the manual tracking burden, with review still needed in patients with many nodules.
- Automated matching worked best with a single nodule and less well with more than five.
- Most AI matching failures were pleural plaques and other non-nodular structures.
- Review AI matching manually in patients with multiple nodules.
- Performance has not been tested prospectively in Indian or other diverse populations.
Why it matters
Removes a labour-intensive step that limits how many screening scans a programme can read.
The statistics, in plain English
Of every 100 persisting nodules, about 84 were matched automatically and fewer than 2 true nodules needed a person to correct them.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free