Below 8 mm there are not enough visual cues.
Follow-up guidelines are built on size and growth. In the 6–8 mm range they recommend interval surveillance or nothing at all, because at that size very few nodules show visual differences a reader can act on. The nodule is not low risk. Its risk is unknown.
In our clearance reader study, unaided radiologist accuracy on nodules smaller than 8 mm was an AUC of 0.459 — below chance. Readers were asked to assign nodules a malignancy score. The AUC measures how well that score predicts diagnosis, on a scale from 0 to 1. In a challenge cohort of size-matched benign and cancerous nodules, reader accuracy was worse than a coin flip (0.5) in nodules smaller than 8 mm. Visual assessment gave them no reliable signal at all.
The improvement is largest exactly where readers have least to go on.
Reader accuracy with and without mSI™, by nodule size, from the FDA clearance reader study.
Cases in this study were deliberately size-matched between benign and malignant nodules, removing the cue readers depend on most. In this curated cohort designed to challenge readers, baseline reader accuracy was close to random chance. On those same smallest nodules, readers with mSI™ ranked correctly closer to three times in four. All ten radiologists improved.
What this means across a thousand nodules.
Most incidental nodules are small, which means most incidental nodules sit in the range where guidelines offer the least.
About these figures. These numbers are based on RevealDx meta-analyses across published retrospective studies. It is offered as a framework for thinking about the size of the opportunity in a nodule program, not as a prediction of results at any given institution. RevealAI-Lung is cleared for nodules 6–15 mm, so within the sub-8 mm group it applies to nodules from 6 mm upward.
Three ways to put a number on a 6 mm nodule.
Tools that estimate malignancy risk draw on different information.
| What it measures | What it can say about a 6 mm nodule |
|---|---|
| Patient demographics and risk factors — age, sex, smoking history, nodule count, lobe location | The same thing it says about every nodule in that patient. Nothing specific to this lesion. |
| Visible morphology — size, calcification, spiculation, margin | May provide some assistance, but the utility is inconsistent, and each characteristic has counter-examples. |
| Sub-visual voxel-level features, compared against a reference set of confirmed outcomes | A lesion-specific assessment that does not depend on the nodule being large enough to look at. |
CPT Appendix S, the AMA’s taxonomy for AI medical services, defines an augmentative output as a score or index that is “more than a summation of quantitative data inputs.” The mSI™ is an augmentative output, derived from voxel-level analysis of the tissue itself. That is what the 0721T and 0722T descriptor names: quantitative CT tissue characterization. AMA CPT Appendix S →
See what mSI™ returns on small nodules from your own reading list.
Twenty minutes with our clinical team. We will show you where the score changes a recommendation in the 6–8 mm range, and where it does not.