Can AI Accurately Track Incidental CT Scan Findings?

Clinicians are increasingly using low-dose chest imaging to detect pulmonary nodules early. However, these scans also contain a wealth of additional health information that doctors often overlook. Using a lung cancer screening CT scan allows advanced software to identify other critical findings automatically. Consequently, we must understand if these artificial intelligence measurements remain consistent over time. A recent landmark study investigated this repeatability using data from the prestigious NELSON trial.

Evaluating AI Repeatability on lung cancer screening CT

Specifically, researchers applied specialized chest software to low-dose, non-contrast scans. The automated tool measured aorta diameters, coronary artery calcium volume, vertebral height, and emphysema percentage [1]. Subsequently, the team calculated the severity of these conditions based on the automated measurements. The study included 1,436 subjects who underwent two scans within a maximum interval of 120 days. On average, the patients had a mean age of approximately sixty years, and most were male. Moreover, more than half of the participants were active smokers during the study period.

Clinical Results and Findings

The AI software demonstrated highly stable measurements for most incidental findings across the scan pairs. For example, the mean absolute difference for aorta diameters ranged between a mere 0.7 and 1.5 millimeters. Meanwhile, the median absolute difference for emphysema severity was remarkably low at just 0.8 percent. In addition, aorta diameter measurements showed good to excellent repeatability with high correlation coefficients. Therefore, clinicians can confidently rely on these automated assessments for longitudinal patient tracking. However, some parameters like coronary calcium volume exhibited slightly higher variability between the consecutive scans.

Implications for Clinical Practice

Ultimately, these findings suggest that modern software can significantly streamline the reporting of incidental findings. Consequently, radiologists in India and globally can save valuable time during routine screenings. Furthermore, automated quantification helps track cardiovascular risks and bone health without extra radiation. As a result, patients receive a more comprehensive health evaluation from a single scan. Therefore, incorporating these automated tools into clinical workflows may improve overall preventive care.

Frequently Asked Questions

Q1: What are the main incidental findings quantified by AI in this study?

The software evaluated several markers, including aorta diameters, coronary calcium, vertebral height, radiodensity, and emphysema percentage.

Q2: How repeatable are the AI measurements on a lung cancer screening CT?

Most measurements showed highly stable results. Specifically, aorta diameters and vertebral heights exhibited excellent repeatability. Meanwhile, emphysema and coronary calcium showed slightly higher variability.

Q3: Why is repeatability important for clinical practice?

Consistent repeatability ensures that any changes observed over time represent actual disease progression rather than software measurement errors. Consequently, doctors can make more reliable treatment decisions.

References

1. Bunk S et al. Repeatability of AI-quantified incidental findings on lung cancer screening CT scans in the NELSON trial. Eur Radiol. 2026 Jun 17. doi: 10.1007/s00330-026-12669-3. PMID: 42310036.
2. Hamelink I et al. Repeatability of AI-based, automatic measurement of vertebral and cardiovascular imaging biomarkers in low-dose chest CT: the ImaLife cohort. Eur Radiol. 2025 Jul;35(7):3833-3841. doi: 10.1007/s00330-024-11328-9. PMID: 39779514.
3. Marques Dos Santos F et al. Comparing artificial intelligence ai-rad companion chest CT vs experts on emphysema, lung nodules, thoracic aorta and thoracic spine. Acta Radiol Port. 2024. doi: 10.25748/ARP.29177.