Aidoc carries three marks because it satisfies all four criteria, but the strongest of them is real-world scale. The CARE Multi-Triage CT Body module cleared the FDA on 21 January 2026 (K252970, fourteen abdominal indications); the platform has CE-MDR and TGA registrations on top of more than twenty FDA clearances. The published Zia et al. prospective study at a UK trauma centre (Scientific Reports, 2022, n = 1,446) reports 85.7% sensitivity and 96.8% specificity for intracranial haemorrhage detection, and a 17-facility academic health-system real-world evaluation in npj Digital Medicine (2025, n = 101,944) documents performance degradation on subacute and small haemorrhages — the kind of honest external-validation finding an editor wants to see in a three-mark device.
The AurisMark Rubric
Version 1.0 · published 2026-05-10
The AurisMark is MedScan's editorial quality signal for AI-enabled medical devices — three marks, two marks, one mark, or unrated. Marks are assigned by editorial judgment against the criteria below; they are not a formula and they are not for sale. This page is the public methodology home: the criteria, the principles, six worked examples, and the primary sources our editors consult when assigning a mark.
01What the AurisMark is
The AurisMark borrows its register from the Michelin Guide — three tiers, distinct meaning, no inflation. A device with three marks is not 50% better than a device with two; it is categorically different along four axes: regulatory footprint, evidence grade, real-world deployment, and safety record. A device that does not meet two-mark criteria does not get two marks. Reviewers err toward fewer marks rather than more — the authority of the rating depends on it.
The AurisMark sits alongside, not in place of, the Clinical Pulse (a quantitative composite of verified clinician reviews) and the EvidenceGrade badge (which describes a single study's design). The AurisMark is the editorial verdict; the other signals are inputs to that verdict and remain visible on every device page so a reader can disagree with our editors and reach their own conclusion.
02Why a rubric, not a formula
A formula is auditable; an editorial rubric is defensible. We chose the second because the questions an AurisMark answers are not weighted-sum questions. A device with an FDA De Novo, no CE mark, and a single 900-patient pivotal trial is not arithmetically comparable to a device with three 510(k) clearances, a CE mark, and a retrospective registry of 50,000 patients. Both can be excellent; reducing them to a single weighted score collapses the editorial reasoning that matters most to a clinician considering procurement.
The model we follow is the NICE Methods Manual: explicit criteria, explicit principles, explicit re-evaluation cadence. When the criteria need to change, the manual gets a new version number and the old version stays archived at its original URL. Future revisions of this rubric will ship at /methodology/auris-mark-v1-1, /methodology/auris-mark-v2, and so on. The v1.0 criteria remain quotable and citable here forever.
03The four bands
The criteria below are the v1.0 rubric, reproduced verbatim from prompts/auris-mark-rubric.md. They are the test our editors apply to every device that receives a mark.
Awarded only for devices that satisfy all four conditions:
- Global regulatory footprint — FDA clearance plus CE marking plus at least one of UKCA, TGA, Health Canada, or PMDA.
- RCT-grade evidence published in a peer-reviewed venue.
- Significant real-world deployment data — at least twenty-four months in market and at least one hundred installation sites, or an equivalent published cohort.
- No unresolved safety alerts or recalls.
Awarded for devices with at least:
- Substantial regional approval — FDA clearance, or CE-MDR / IVDR with documented broad uptake.
- Prospective-grade or better clinical evidence.
- Manageable safety record — resolved alerts are acceptable.
Awarded for devices with at least:
- One regulatory approval — FDA, CE, or an equivalent competent authority.
- Some published evidence on file at any grade.
Devices that have not yet been editorially evaluated render without an AurisMark. The lean detail page still surfaces the FDA submission record, the manufacturer disclosure, and any clinician reviews. Lean entries are promoted to a mark when an editor has reviewed the device against the criteria above.
04Six worked examples
Six devices in our catalogue currently carry the AurisMark. Each illustrates one dimension of the rubric — chosen so that a reader can read the editorial verdict back to its primary sources without going through us.
HeartFlow earns three marks on evidence depth. The pivotal NXT study (Nørgaard et al., JACC, 2014, n = 254 across four countries) was the De Novo dataset (DEN130045); subsequent randomised trials — PRECISE, FORECAST — and the 90,553-patient NHS England FISH&CHIPS cohort published in Nature Medicine (May 2025) place FFRCT in a different evidence class from most AI cardiology devices. PRECISE reported a four-fold reduction in unnecessary invasive catheterisation; FORECAST reported a 22% reduction in invasive testing overall. The Category I CPT code that took effect on 1 January 2024 confirms reimbursement at standard-of-care level.
Paige earns three marks for being the first AI product authorised by the FDA for use in digital pathology — the De Novo classification (DEN200080) on 21 September 2021 created an entirely new product code (QMX) that subsequent pathology AI submissions reference as predicate. Raciti et al. in the Archives of Pathology & Laboratory Medicine (2023) reported a 7.9-percentage-point sensitivity gain among eighteen pathologists across 610 whole-slide images; the CE-IVDR mark and UKCA were obtained on the same evidence package. A first-in-class De Novo with its own product code, published prospective evidence, and EU + UK regulatory parity meets the rubric in a category of one.
LumineticsCore (formerly IDx-DR) is the first autonomous diagnostic AI authorised in any field of medicine — the FDA De Novo (DEN180001) on 11 April 2018 explicitly reads no human in the loop. The pivotal trial (Abràmoff et al., npj Digital Medicine, 2018, ten sites, n = 900) reported 87.2% sensitivity and 90.7% specificity for referable diabetic retinopathy. The ACCESS randomised trial (Nature Communications, 2024) showed an autonomous-AI screening completion rate that materially outperformed standard care. Three marks on rubric grounds and on regulatory novelty — the bar for autonomous AI in clinical medicine starts here.
Caption Health's sonographer-guidance system was the first of its kind — FDA De Novo (DEN190040) on 7 February 2020 under Breakthrough Device designation. Narang et al. in JAMA Cardiology(2021, n = 240) reported 92.5% diagnostic-quality TTE acquisitions by nurses with no prior echo training. Subsequent acquisition by GE HealthCare integrated the algorithm into the Venue Family and Vscan Air SL — the rubric's real-world deployment criterion is met not by a standalone install base but by OEM channel reach across four cleared markets (US, EU, Canada, Australia). A useful test case for procurement officers asking what real-world deployment scale looks like outside a hospital IT contract.
CADDIE earns three marks on the freshness of the evidence base. The EAGLE randomised controlled trial (n = 841, published 2025) is one of the first multi-centre RCTs of an AI computer-aided detection system in colonoscopy and reported a meaningful adenoma-detection-rate improvement alongside secondary endpoints on large adenomas, sessile serrated lesions, and non-polypoidal adenomas. The FDA 510(k) cleared the device on 5 September 2024 (K240044); CE-MDR followed in October 2024. The device is younger than every other entry above; the rubric's freshness principle (re-evaluate annually, marks may move down as well as up) applies most acutely here.
05Editorial principles
Four principles govern every assignment. They are the non-negotiable text from prompts/auris-mark-rubric.md and they take precedence over any specific criterion above.
- Honesty over generosity.A device that doesn't meet two-mark criteria should not get two marks. The mark's authority depends on rigor.
- Public sources only. Marks are assigned from publicly verifiable evidence — not vendor briefings, not unpublished trials. If a vendor offers data we cannot cite, we decline it.
- Re-evaluate annually. Marks may move down as well as up. New safety alerts, expired approvals, or superseded clinical evidence can downgrade. Every device record carries a
last_reviewed_atstamp. - No appeals process. Vendors do not pay for marks, and they cannot influence them. This is the brand moat; we will lose customers before we will trade it.
06Versioning
The current version is v1.0, published 2026-05-10. The criteria above are the immutable v1.0 text and will remain at this URL — even after a future revision lands.
When the criteria change, we publish a new version at a new URL (/methodology/auris-mark-v1-1, /methodology/auris-mark-v2) with a public diff against the prior version. Devices are then re-evaluated under the new rubric, and individual mark changes are documented in each device's revision history. Citing this v1.0 page in a paper, a guideline, or a procurement document is therefore durable: the criteria the citation refers to will not silently change underneath it.
07What the AurisMark is not
The AurisMark is an editorial signal about a device's standing in its category. It does not, on its own:
- Replace the Clinical Pulse. The Clinical Pulse aggregates verified-clinician reviews into a quantitative 0–100 composite. The AurisMark is editorial; the Clinical Pulse is empirical. They answer different questions and a reader benefits from seeing both.
- Recommend procurement. A three-mark device is not necessarily the right device for a given hospital, budget, integration footprint, or patient mix. The mark says the device meets a high editorial bar; institutional fit is its own question.
- Rate clinical evidence in isolation.The EvidenceGrade badge on every device card describes a single study's design (RCT, prospective, retrospective, expert opinion). The AurisMark integrates evidence with regulatory footprint, deployment scale, and safety record; it is not a substitute for reading the underlying trials.
- Capture rare-event safety signals. Post-market surveillance issues live in the Safety Alerts section of each device page and are sourced from regulatory databases, not from editorial judgement.
08Primary sources
The criteria above are applied against publicly verifiable documents. The list below is the canonical source set; per-device citations live on each device's detail page.
- FDAArtificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices — the canonical FDA list of authorised AI/ML medical devices.
- FDA510(k) Premarket Notification database — searchable index of clearances, including 510(k) summaries.
- FDADe Novo Classification database — first-in-class authorisations referenced in several worked examples above.
- European CommissionEUDAMED — European Database on Medical Devices — CE-MDR and IVDR registration source of record.
- EU AI ActRegulation (EU) 2024/1689 — the Artificial Intelligence Act — consolidated text governing high-risk AI systems including medical devices.
- NICEDeveloping NICE guidelines: the manual (PMG20) — the methodology-versioning analogue we follow for this rubric.
- IMDRFSoftware as a Medical Device — Risk Categorization Framework (N12) — international risk-categorisation reference for SaMD.
Comments on the rubric — statistical, ethical, or practical — are welcome at methodology@medtekki.no. Substantive feedback is logged with the next version's release notes.
AurisMark Rubric v1.0 — published 2026-05-10 — MEDTEK KI AS · Org.nr 937 565 704 · Stavanger, Norway. Published under CC BY 4.0.