Validated in over 1,500 patients. mSI™ improved 10 of 10 radiologists in over 100 cases.
RevealAI-Lung accuracy has been validated in more than 1,500 patients — screening and incidental, low-dose and contrast-enhanced — with peer-reviewed results in the Journal of the American College of Radiology and Clinical Radiology, and external validation above 0.8 AUC in three independent populations in the US, Canada and the UK. Separately, its effect on practicing radiologists was measured directly in the multi-reader, multi-case study supporting FDA clearance: all ten readers improved, across experience levels, on cases matched so that size gave no help.
Every figure below is given with its limits stated — what each study was designed to test, where a result depends on that design, and what the cleared label does not cover. "Independent Literature" tags indicate third party studies and literature that were published independently of RevealDx.
In a UK cohort, mSI™ would have accelerated cancer follow-up by 44%.
We retrospectively analyzed 100 cancer cases that were followed from incidental lung nodule CT detection to diagnosis. We asked whether mSI™ could have improved patient follow-up compared to standard of care if it had been used in the first scan. BTS guidelines alone would have recommended interval scan and delayed cancer diagnosis in 45/100 malignant nodules. Using mSI™ to upgrade risk would have accelerated follow-up in 20/45 of those cases, and never delayed follow-up more than standard of care.
Retrospective single-center case-control study, UK district general hospital, BTS 2015 guidelines. Based on retrospective, standardized application of BTS guidelines, rather than observed clinical practice. Follow-up observed in practice did not necessarily strictly follow BTS guidelines, but retrospective analysis showed that follow-up was broadly concordant with guidelines. Routine CT: 66% contrast-enhanced, 19 convolution kernels, all four major vendors. Cohort enriched for higher-risk nodules. Includes lung nodules outside the US indication of 6–15 mm. Duerden et al., Clinical Radiology, 2026. doi:10.1016/j.crad.2025.107135
Clearance reader study (K251769)
A multi-reader multi-case study of ten radiologists reading with and without the Malignancy Similarity Index (mSI™). 108 cases were deliberately size-matched between benign and malignant nodules, removing the cue readers depend on most, which is why baseline accuracy sits near chance.
| Measure | Without mSI™ | With mSI™ |
|---|---|---|
| Mean reader AUC | 0.538 | 0.719 |
| Readers improved | 10 of 10, range 0.11 to 0.26 AUC | |
| Statistical significance | p<0.0001 | |
Improvement held across reader experience, patient age, sex and race, scan contrast and manufacturer, and nodule size, lobe and opacity. Sensitivity rose 14 points and specificity 12 points at a pre-specified threshold; those figures depend on the study sample and should not be read as population statistics.
Where the improvement is largest.
Reader AUC by nodule size. The gain is widest on the smallest nodules — the ones where unaided visual assessment has the least.
Upstaging is most common in the smallest tumors.
In 390 screen-detected cancers 30 mm or smaller at referral, 43% increased in tumor stage before treatment. The smallest upstaged most often: 56% of those 10 mm or under, against 38% at 10–20 mm and 40% at 20–30 mm. Upstaging was associated with higher mortality. The authors attribute the pattern to diagnostic uncertainty below 10 mm, where lesions are commonly monitored for growth.
Mullin ML, Verghese P, Khaw CR, et al. Upstaging of screen-detected lung cancers during diagnostic assessment. Thorax 2026;81:894–902. doi:10.1136/thorax-2025-224006. SUMMIT prospective cohort; 390 cancers staged cT1a–c at referral (clinical tumor stage cT1a is 10 mm or under, cT1b 10–20 mm, cT1c 20–30 mm); 165 (43%) upstaged, p=0.01 across categories; adjusted mortality HR 1.68 (95% CI 1.13–2.51), p=0.01. Upstaging measured from referral to first treatment. The authors report an association with time to treatment and state that a causal link cannot be concluded.
Growth takes 400 to 800 days to appear in subsolid malignancy.
Waiting for nodules to grow means waiting for cancer to progress. The ACCP guideline reports growth rates of 400 to 800 days for malignant subsolid nodules, with doubling times as long as 1,500 days. The mSI™ is returned on the lung nodule at the first scan. With mSI™, more patients who need attention are identified on day one, not after the interval.
Gould MK, Donington J, Lynch WR, et al. Evaluation of Individuals With Pulmonary Nodules: When Is It Lung Cancer? Diagnosis and Management of Lung Cancer, 3rd ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines. CHEST 2013;143(5 Suppl):e93S–e120S. Growth rates cited are for subsolid malignant nodules; solid malignant nodules usually double in 20 to 400 days. In one screening series of 13 subsolid malignant nodules, mean time for growth to become detectable was 715 days by diameter. Management recommendations in the 3rd edition have since been revised by Fleischner 2017 and Lung-RADS 2022.
External validation
Standalone algorithm performance in independent populations, without reader interpretation, and in addition to the NLST cross-validation cohort.
| Populations | Patients | Result |
|---|---|---|
| Three independent cohorts — United States, Canada, United Kingdom | 675, of whom 276 had cancer | AUC above 0.8 in each |
Trained against confirmed outcomes in the National Lung Screening Trial. Validated across more than 1,500 patients, including NLST cross-validation and external validation, spanning screening, incidental, low-dose and IV-contrast studies — the range of scans radiologists actually read.
Publications
Duerden et al. — Clinical Radiology, UK validation
Adams et al. — Journal of the American College of Radiology, US multi-site validation
Autrusseau et al. — Diagnostic and Interventional Imaging, full-dose and ultra-low-dose CT concordance
Full text and study PDFs available on request.
Questions a study summary will not answer?
Our clinical team will walk through methodology, cohorts and limitations directly.