RapidAI is a modular stroke imaging platform. Its modules fall into three roles.
Triage and notification (alerts the team, no diagnosis)
Diagnostic aid (read together with the radiologist)
Treatment selection (adults only)
FDA: all US clearances below are 510(k), Class II, applicant iSchemaView, Inc., Golden, CO.
| Module | Reference | Date | Status |
|---|---|---|---|
| United States — FDA 510(k) | |||
| Rapid ICH | K193087 QAS, 21 CFR 892.2080 |
31 Mar 2020 | To check (note 1) |
| Rapid ASPECTS | K200760 POK, 21 CFR 892.2060 |
26 Jun 2020 | Confirmed (FDA database) |
| Rapid LVO | K200941 QAS, 21 CFR 892.2080 |
9 Jul 2020 | Confirmed (FDA database) |
| Rapid LVO (later version) | K221248 | 31 May 2022 | Confirmed (FDA database) |
| Rapid NCCT Stroke | K222884 QAS, 21 CFR 892.2080 |
2 Mar 2023 | Confirmed (FDA database) |
| AngioFlow (perfusion in the angiography suite) | 510(k), number not checked | Announced May 2024 | Vendor-claimed |
| Rapid DeltaFuse, Rapid LMVO, Rapid MLS, Rapid OH, Rapid Aortic | Five clearances, numbers not checked | Announced 25 Nov 2025 | Vendor-claimed |
| European Union / Europe | |||
| Rapid ASPECTS | Described as CE-marked at European launch (ESOC) Likely MDD (interpretation, to check; note 2) |
May 2018 | Vendor-claimed |
| AngioFlow | Europe (unspecified): vendor CEO, “already registered and used in Japan and Europe” | – | Vendor-claimed (note 3) |
| United Kingdom | |||
| RapidAI | NICE HTG708 | 2024 | Conditional NHS use (note 4) |
Notes
Regulatory submission study (Rapid ASPECTS, K200760): concurrent-read crossover study, 50 cases, 8 readers of mixed specialties; software assistance added 0.425 regions of agreement per scan (95% CI 0.11-0.74); non-neuroradiologist agreement rose from 73.6% to 79.8% (p<0.0001). Summary from a third-party 510(k) database, not re-read on the FDA PDF. Vendor-claimed performance (Rapid LVO, clearance press release, 13 August 2020): 97% sensitivity, 96% specificity; press-release figures, not peer-reviewed. "DUEL" study: Sachdev H, Hudson A, Ong K, Marklein S, Flores M, AJNR, published online 14 May 2026, DOI 10.3174/ajnr.A9418. Single comprehensive stroke centre (Good Samaritan Hospital, HCA, San Jose, USA), 1,589 consecutive stroke alerts, 1,523 analysed, 147 confirmed LVOs. Rapid LVO 98% sensitivity (144/147) and about 94% specificity versus Viz.ai 73.5% (108/147) and about 91% (p<0.0001 and p=0.004). Promoted by the vendor; funding and conflict-of-interest statements not verified on the full article.
Lambert J, Demeestere J, et al. (University Hospitals Leuven / KU Leuven, Belgium), AJNR 2023;44(8):894-900, DOI 10.3174/ajnr.A7956: retrospective single-centre study, 174 thrombectomy-treated anterior-circulation stroke patients (175 non-contrast CTs), 5 raters. RAPID ASPECTS alone: weighted kappa 0.59 versus the criterion standard, similar to human raters (0.38-0.76). With software assistance, accuracy improved (AUC 0.72 to 0.82, p=0.004) and inter-rater agreement improved (Fleiss kappa 0.15 to 0.23 overall; 0.39 to 0.55 for dichotomized scores); for the dichotomized thrombectomy threshold, most individual rater improvements were not significant. Author-stated limits: thrombectomy-only population (low ASPECTS under-represented), criterion-standard readers had access to the software output, one vendor tested, heterogeneous referral CT protocols.
CDA-AMC (Canada’s Drug Agency) HTA, expert panel meeting 12 September 2024, recommendation December 2024: 2 cohort and 11 diagnostic accuracy studies, only 3 testing RapidAI as intended (alongside clinicians); standalone LVO sensitivity about 62-96% and specificity about 65-98% across 10 studies; ICH 92% sensitivity (95% CI 78-98) and 100% specificity (95% CI 98-100) from one low-certainty study; no evidence on mortality, length of stay, quality of life, patient harms or cost-effectiveness; all studies in centres with stroke expertise. Where implemented: use only alongside clinician interpretation and generate evidence; for new implementation: no recommendation for or against.
NICE HTG708 (UK, 2024): observational time-to-treatment studies including RapidAI judged at high risk of bias; use allowed only with healthcare professional review while evidence is generated.
Vendor claim (2026): used in 2,500+ hospitals globally. Earlier vendor figure (March 2021): over 1,800 hospitals in over 60 countries. Both are vendor-reported and not independently audited. Named sites confirmed by third parties are NHS trusts in the West Midlands, UK (Queen Elizabeth Hospital Birmingham as hub, Hereford County Hospital, Sandwell and City Hospitals, Worcestershire Royal Hospital, 2023-2024). No named routine deployment in Belgium or another EU country was found.
Vendor-claimed: standard integration via DICOM, HL7 and FHIR plus custom APIs; DICOM-based worklist prioritization or integration with worklist providers; AI outputs surfaced alongside native images in PACS; integration with RIS and EHR systems; described as vendor-agnostic and built for hybrid environments (“Edge Cloud” option). No DICOM conformance statement, HL7 version or FHIR resource list was found in public sources; confirm these, the supported PACS/RIS/EHR vendors and the cloud/on-premise architecture with the vendor before deployment.
– Evidence certainty rated low to very low by an independent HTA body (CDA-AMC, 2024); NICE (2024) allows NHS use only while more evidence is generated.
– Standalone LVO detection varies widely across studies (sensitivity about 62-96%, specificity about 65-98%); most accuracy studies tested the software standalone rather than as intended alongside clinicians.
– No independent evidence on mortality, length of stay, quality of life, patient harms or cost-effectiveness.
– Evidence comes from centres with stroke expertise; performance in smaller or less-resourced hospitals is untested.
– No public data on training-data representativeness or subgroup performance (age, sex, ethnicity); risk of automation bias noted by CDA-AMC.
– Triage modules (LVO, ICH, NCCT Stroke) are notification tools, not diagnostic devices: absence of an alert does not rule out the finding.
– Rapid ASPECTS is a concurrent-read aid only; independent agreement with expert consensus is moderate (weighted kappa 0.59).
– Current EU MDR certificate, risk class and notified body not publicly confirmed; EU data hosting not documented.
– No named routine deployment in Belgium or another EU-27 country found; European evidence of use is mainly from NHS England.
Artificial intelligence aids faster stroke diagnosisHospital — Worcestershire Acute Hospitals NHS Trust, AI software helping stroke patients, 2024 https://www.worcsacute.nhs.uk/artificial-intelligence-software-helping-stroke-patients-at-worcestershire-acute-hospitals-nhs-trust/ Regulation — EU, Regulation (EU) 2024/1689 (AI Act), EUR-Lex, July 2024 https://eur-lex.europa.eu/eli/reg/2024/1689/oj
This profile is not medical advice or an endorsement. Hospitals should perform their own due diligence (regulatory compliance, GDPR, cybersecurity, integration, local clinical validation). How we evaluate
Every profile documents the precise indication, regulatory status, vendor-supported and independent evidence, real-world deployment, limits and dated sources.
A independent systematic review or prospective multicentre study · B at least one independent peer-reviewed study · C peer-reviewed, vendor-supported · D vendor claims only.
Vendors, authors and clinicians can request a correction. Corrections are dated on the profile.