Radiology · Chest-radiograph abnormality detection and triage
Lunit INSIGHT CXR
Lunit Inc.
Lunit INSIGHT CXR is among the most widely trialled chest-radiograph detection tools, but its record rewards a careful reading. The only US clearance is a narrow triage notification, not the flagship detector; and its most pragmatic effectiveness trial — 3,576 emergency patients — found no benefit over conventional reading.
Performance Metrics
Clinical Evidence
The evidence base is unusually large for a chest-radiograph product and, in places, unusually candid. Two randomised trials frame it. In a health- screening population, an independent single-centre [[rct]] of 10,476 participants (Radiology 2023) embedded INSIGHT CXR v2.0.2.0 into routine reading and found it improved detection of actionable Lung-RADS category 4 nodules — 0.59 percent versus 0.25 percent, p = 0.008 — at similar false- referral rates. The counterweight is more consequential: a pragmatic [[rct]] of 3,576 emergency-department patients with acute respiratory symptoms (Korean J Radiol 2023) found AI assistance did not improve [[sensitivity]] for acute thoracic disease (67.2 versus 66.0 percent) or the false-positive rate (19.3 versus 18.5 percent), and moved none of the downstream outcomes — CT use, antibiotic use, length of stay, 30-day revisit. The authors noted the contrast with their own earlier retrospective work, a plain illustration of how reader-study gains can fail to survive contact with everyday practice. In tuberculosis screening the independent record is stronger. A five- algorithm evaluation in Dhaka (Lancet Digit Health 2021) placed INSIGHT CXR v4.9.0 at an area under the curve of 0.886, mid-field among peers and above radiologists, and a prospective household-contact cohort in South Africa (Clin Infect Dis 2025) reported an area under the curve of 0.91 for INSIGHT v3.1.4.111, ahead of CAD4TB v7.0 (0.87) and qXR v3.0.0 (0.88). Read as a body, the work supports the product as a capable detector and TB-screening aid while cautioning that its measured benefit is setting-dependent — clear in screening, absent in the acute-care trial.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Hwang EJ, et al. (Seoul National University, pragmatic RCT) | RCTRCT | 3,576 | 67.2% (AI-assisted) vs 66.0% (conventional) | false-positive 19.3% vs 18.5% | — | Korean J Radiol 2023;24(3):259-270; no significant difference; AI-CAD did not improve sensitivity, false-positive rate, CT use, antibiotic use, length of stay or 30-day revisit; PMID 36788769 |
| Independent single-centre RCT, health-screening cohort | RCTRCT | 10,476 | Lung-RADS 4 detection 0.59% (AI) vs 0.25% (non-AI) | similar false-referral rates in both arms | — | Radiology 2023; p=0.008; INSIGHT CXR v2.0.2.0 embedded in routine reading; PMID 36749213 |
| Qin ZZ, et al. (independent, Dhaka, Bangladesh) | RetrospectiveRetrospective | 23,954 | AUC 0.886 | all five algorithms reduced Xpert tests ~50% at >90% sensitivity | 0.886 | Lancet Digit Health 2021; five-algorithm TB-triage evaluation; INSIGHT CXR v4.9.0; PMID 34446265 |
| Independent prospective household-contact cohort, South Africa | ProspectiveProspective | 483 | AUC 0.91 (0.83–0.99) | vs CAD4TB v7.0 0.87, qXR v3.0.0 0.88 | 0.91 | Clin Infect Dis 2025; rifampicin-resistant-TB household contacts, 5-year follow-up; INSIGHT v3.1.4.111 |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow INSIGHT CXR reads a chest radiograph.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
A frontal chest radiograph arrives from PACS.
INSIGHT CXR receives a digital frontal () chest radiograph from or the imaging gateway. The product is a reading aid for a single study at a time; the CXR4 line can additionally reference a prior study to flag change, but the core output is per-image.
Deployment is server-side and version-locked — the independent literature cites specific releases (v2.0.2.0, v4.9.0, v3.1.4.111), indicating a controlled release lineage rather than a silently updating model.
Input
Chest radiograph
Inference
Per-study
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
K211733
Class II (21 CFR 892.2080, product code QFM)
Safety Record
No recall, field safety notice or regulator safety communication specific to Lunit INSIGHT CXR was found in public reporting as of July 2026, and a search of the FDA MAUDE and enforcement databases at digest mining returned no matches. That FDA channel covers only the narrow triage product (K211733), since the full detection device is not FDA-distributed; the primary EU, UK, Canadian and Korean registers could not be queried directly from this environment, so absence of a public alert is not equivalent to a clean register read. The software is a concurrent or second reader — every finding passes through the reporting radiologist, and the absence of a mark is not an exclusion of disease.
Intended Use & Indications
Lunit INSIGHT CXR analyses digital frontal ([[dicom]]) chest radiographs and marks regions suspicious for a defined panel of thoracic findings, returning an abnormality score per finding and a localisation overlay for the reporting radiologist. It operates as a [[cade]] aid — concurrent or second reader — and does not act autonomously. The US-cleared product is a distinct and narrower device, Lunit INSIGHT CXR Triage (FDA K211733): a computer-assisted triage-and-notification tool that flags studies with pre-specified critical findings (pleural effusion and/or pneumothorax) for prioritisation. It does not provide the diagnostic detection output of the full INSIGHT CXR product and is not a diagnostic device.