Truace
Subscribe

AI matched 283 of 339 lasting lung nodules across scans in a UK screening trial

Study title: Automated artificial intelligence performance for longitudinal pulmonary nodule matching in lung cancer screening

Researchers tested a pulmonary AI system for fully automated longitudinal nodule matching in 361 UK Lung Cancer Screening trial participants who had 3-month follow-up low-dose CT. Using a >=100 mm3 solid-component threshold, the AI found 378 baseline nodules in 181 participants; 39 resolved, and it matched 283 of 339 persisting nodules for an 83.5% success rate, with 91.8% success in single-nodule cases and 72.8% when more than five nodules were present.

Read the study abstract

Accurate longitudinal nodule matching is a critical technical prerequisite for automated growth rate (volume doubling time) assessment in lung cancer screening. This study evaluated an artificial intelligence (AI) pulmonary nodule analysis system in all 361 UK Lung Cancer Screening (UKLS) trial participants who underwent a 3-month follow-up low-dose computed tomography (LDCT) scan. The pulmonary AI independently evaluated these baseline scans using an updated volume threshold (solid component ≥ 100 mm³ per NELSO…

By the Truvace desk1 min read1 sourceEvidence: Moderate · 60/100 — Observational study · n = 361 · 1 independent sourcePeer-reviewedTRV-2026-0933· cite & verify
Automated artificial intelligence performance for longitudinal pulmonary nodule matching in lung cancer screening
Metastatic renal cell carcinoma - CT scan - Case 262 (8559836650) by Yale Rosen from USA. CC BY-SA 2.0 · https://creativecommons.org/licenses/by-sa/2.0
Trace impact readingContested
P 58The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 55The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

The result matters because accurate longitudinal matching is a prerequisite for automated volume doubling time assessment, and the study suggests manual tracking workload could be reduced since only 1.5% of persisting findings were discrete solid nodules needing correction. Uncertainty remains about performance in high-burden scans where most failures were non-nodular pleural plaques, and about generalizability beyond the UKLS cohort without prospective validation in diverse populations.

Main points

  1. Study included all 361 UKLS trial participants who had 3-month follow-up LDCT, with AI evaluating baseline scans using solid component >= 100 mm3 per NELSON 2.0/EUPS protocol.
  2. AI detected 378 baseline nodules >=100 mm3 in 181 participants; 39 resolved by follow-up, leaving 339 persisting nodules for automated matching without manual selection.
  3. Of 56 unmatched findings, 91.1% were non-nodular structures, predominantly pleural plaques at 46.4%.

The gain

Automated pulmonary AI matched persisting lung nodules across 3-month LDCT scans with 83.5% success, reaching 91.8% for single-nodule cases and leaving only 1.5% of persisting findings needing manual correction, indicating potential to reduce manual tracking workload.

The problem

Automated matching failed for 16.5% of persisting findings and performance fell to 72.8% in participants with more than five nodules, with prospective validation in diverse populations still needed.

The rundown

The evaluation used an updated volume threshold of solid component >= 100 mm3 and proceeded to fully automated longitudinal matching without any manual selection to test algorithmic robustness.

Among 361 UKLS participants with 3-month follow-up LDCT, the AI identified 181 participants with 378 baseline nodules, of which 39 had naturally resolved, leaving 339 persisting nodules for matching analysis.

What this doesn’t fix

Performance drops with high nodule burden and generalizability remains unproven, requiring prospective validation in diverse populations.

Sources

  1. Peer-reviewedEuropean Radiology2026-08-29

The debate