Lung nodules can look identical. mSI™ helps reveal the cancer.
Too many lung nodules are just waiting for the next scan.
Lung cancers are diagnosed too late. Benign nodule follow-up burdens doctors and patients. RevealAI-Lung measures digital biomarkers hidden in the CT scan and returns the Malignancy Similarity Index: the mSI™. That helps you separate benign nodules from cancer and focus on the lung nodules that matter.
mSI™: the Malignancy Similarity Index.
The score is a continuous value from 0 to 1, computed from the radiomic features of a single lung nodule and compared against a reference database of nodules whose diagnoses were confirmed over years of follow-up. It is a relative similarity score, not a percentage probability of cancer. Grouping that scale into three bands — low, intermediate and high — aids interpretation.
The reading radiologist interprets the mSI™ alongside nodule size, growth and clinical history, and the score supports shared decision making with the patient.
How the score is produced and read →Lung nodules need better follow-up than waiting for cancer to grow.
With small lung nodules, guidelines often recommend waiting to see if the nodule grows. In 76 screen-detected cancers, the median time to detect an increase in nodule size was 13 months. We retrospectively analyzed over 1,500 diagnosed lung nodule cases, and asked whether mSI™ could have accelerated diagnostic follow-up. Among cancers where guidelines recommend waiting, mSI™ would have accelerated follow-up in 45% from the first scan.
by interval scan
Of 76 screen-detected cancers followed to diagnosis, by linear measurement. Independent study.
with mSI™ (retrospective)
Of the malignant lung nodules that guidelines would have deferred, across retrospective validation cohorts.
Growth: Byrne SC, Hammer MM, American Journal of Roentgenology, doi:10.2214/AJR.22.27869. 76 Lung-RADS 3+ screening nodules later diagnosed as cancer; 5% grew by linear and 7% by volumetric measurement at three months; median time to growth 13 months (linear), 11 months (volumetric). Reclassification: cancers moved by mSI™ from interval surveillance to accelerated follow-up, as a share of those that guideline-based management (Lung-RADS or BTS) would have delayed — 45% pooled, range 33–59%. Adams et al., JACR 2023; Duerden et al., Clinical Radiology 2026; RevealDx data on file.
Three in four lung nodules are too small to characterize by eye.
Of a thousand incidental lung nodules, roughly 750 measure under 8 mm. That is the visual ceiling — the size below which very few lung nodules show visual differences, and follow-up guidance says wait. Not because the nodule is benign, but because there is too little left to see. Those nodules are not reassuring. Their risk is unknown.
Accuracy on lung nodules smaller than 8 mm in our clearance study — below chance.
The same readers. The same lung nodules.
Readers were asked to assign nodules a malignancy score, with and without assistance from mSI™. The AUC reported above measures the ability of that score to correctly predict cancer on a scale from 0 to 1. On a cohort of cancerous and benign nodules size-matched so that size gave no help, unaided reader accuracy was worse than a coin flip on nodules smaller than 8 mm. With mSI™, accuracy rose closer to three times in four. Even when there is zero visual signal, mSI can reveal the cancer.
Where size stops helping →Ten radiologists. Indeterminate and size-matched nodules to challenge the reader. All ten improved.
Cases in our clearance reader study were matched for size between benign and malignant nodules, so size gave the reader no help. That is why baseline accuracy sits near chance.
Mean reader AUC without and with mSI™ — an improvement of 18 points (p<0.0001).
Every radiologist in the study improved, by between 0.11 and 0.26 AUC.
Who is this for.
RevealAI-Lung touches radiology, independent practice, lung nodule programs and IT. Each page below answers the questions that role actually asks.
“Will this change what I put in the impression — and will it slow me down?”
RevealAI-Lung runs inside your existing PACS viewer, with no second application to open. The score returns in seconds, on a nodule you select or one your detection tool has already flagged. The mSI™ becomes a line in your impression, alongside size and morphology.
How it works →“Are the economics favorable for free standing imaging centers”
Yes, and the economics may even be better. A group that bills for its own CT equipment bills 0721T at place of service 11 and collects the global rate, meaning the technical and professional components together. A group reading on hospital-owned equipment collects the professional component only, and the technical side goes to the facility.
Group economics →“Which patients in this queue need us first?”
The mSI™ travels with the radiology report into the system. A lung nodule tracking program or EHR can sort and provide alerts based on the mSI score. This organizes by risk rather than just date or nodule size.
Lung nodule programs →“What exactly are you putting on my network?”
Docker images inside your own environment — on-premise or in your own cloud. No vendor-operated cloud, no GPU server, two open ports, PHI encrypted at rest with a site-specific key.
Deployment and IT →Delivered through the platforms your department already chose.
RevealAI-Lung is cleared to run inside an existing PACS, and can consume the output of a third-party detection system. It is built to be embedded in the platforms a department already runs.
See RevealAI-Lung on a chest CT.
Twenty minutes with our clinical team. We will show you where the score changes a recommendation, and where it does not.