⟷ drag to compare ⟷

aView · LCS

Axial · Slice 147/320 · W:1500 L:-600

LungAI v2.4

~10 mm

Lung-RADS

4X

Very Suspicious

Manual caliper · rounded · category only · ~5 to 10 min

Malignancy Risk

87/100

High Risk

+1.8 mm vs prior

LungAI · Nodule #1

TypeSolid

Size10.2 × 8.4 mm

Volume384 mm³

Density+28 HU

MarginSpiculated

LocationLUL · periph

VDT240 d

Drag to compare

Workflow

Split view

Read time

5 to 10 min / ~2 min

Output

Category + Score

Billable

Standard fees + CPT 0721T · $650

Drag the divider to compare a manual Lung-RADS read with LungAI CADx on the same scan.

AI-Powered Lung Cancer Screening + CADx Diagnostics

Lung-RADS Tells You the Category.
LungAI Tells You the Probability.

The first reimbursed CADx that scores every nodule 0 to 100, with automatic prior-scan comparison, embedded directly inside the CADe viewer your radiologist already uses.

0.96

AUROC on NLST
vs Lung-RADS 0.88

100M

Parameter
Vision Transformer

$650/scan

Reimbursable
under CPT 0721T

Request a Demo See the Technology

The Problem

Up to 4 in 10 Lung Cancer Diagnoses are Wrong

Lung-RADS v2022 depends entirely on human judgement, a manual rule-based scoring system with no AI and no patient personalization.

How Lung-RADS Works Today

1

Manual chart review by radiologist

2

Size-based rules only (no shape or texture AI)

3

Subjective reading, 5-10 min per scan

4

Single snapshot with no longitudinal tracking

5

Result: blanket biopsy referrals or missed cancers

~10%

False Negatives

Real cancers missed, caught only when it's too late.

Up to 40%

False Positives

Unnecessary biopsies, procedures, and patient anxiety.

30%

Inter-Reader Variability

Two radiologists reading the same scan will disagree on the risk category nearly a third of the time.

A radiologist reading 8,000 lung scans per year will incorrectly categorize roughly 2,400 of them, with no AI safety net.

The Solution

AI That Sees What Lung-RADS Misses

LungAI integrates directly into your existing CADe workflow to deliver a continuous 0–100 malignancy risk score per nodule, backed by a 100-million-parameter model trained on 100,000 CT scans and reimbursable under CPT 0721T.

Architecture

Three-Component AI Model

A 3D Vision Transformer nodule encoder, a Sybil-derived whole-scan encoder, and a dense decoder that fuses both with patient demographics and prior scan data to output a per-nodule malignancy probability.

0.96 AUROC

vs Lung-RADS 0.88 on NLST held-out set

Clinical Performance

Fewer Misses. Fewer False Alarms.

At the high-sensitivity threshold, LungAI achieves 94% sensitivity and ~88% specificity — a 60-80% reduction in missed cancers and a 33% reduction in unnecessary biopsies versus standard reads.

94% Sensitivity

~88% Specificity · 1.5–3 min read time

Integration

Zero New Clicks. Built for Your Workflow.

LungAI scores appear directly in your existing CADe viewer. No new workstation, no new login, no IT project. One integration with a CADe partner unlocks their entire hospital network. Post-FDA clearance, every screened patient generates a CPT 0721T billing event at ~$650 per scan.

$650 per scan

CPT 0721T · post-FDA clearance

Pre-FDA Clearance · NLST held-out retrospective study · Results may differ in prospective clinical use

The Platform

One Platform. Full Funnel.

From patient identification to reimbursable diagnosis, Oatmeal Health covers the entire lung cancer screening journey.

1

🧠

Identify

AI Patient Identification

Our AI analyzes EHR data: notes, labs, imaging, billing codes, to predict patients with 20+ pack-year smoking history, even when undocumented. 9x faster than manual chart review with 0.97 AUROC.

2

🩺

Navigate

Care Navigation + SDM Visit

AI navigators help FQHC nurse practitioners conduct shared-decision making visits via telehealth. We help educate patients and drive a 91% scan agreement rate, feeding the diagnostic pipeline.

3

🎯

Diagnose

LungAI CADx: Malignancy Risk

Our 100M-parameter Vision Transformer delivers automated malignancy risk assessments through native integration with PACS/CADe vendors. Nodule-level inference with longitudinal tracking.

See How it Works →

Who Benefits