RapidAI

The core modules are built on the models, and the platform sold around them carries value of its own. RapidAI's cleared modules for stroke, neurocritical care, aneurysm, aortic and pulmonary embolism imaging are models that do not exist without the algorithms.

What a health system buys, though, is the Rapid Enterprise Platform: Rapid EdgeIQ orchestration that routes studies to the right AI and results to the right care teams, Navigator Pro for worklist prioritization, in-PACS communication and automated reporting, the Rapid mobile app for care team collaboration, data and analytics on utilization and impact, and a channel for partner and hospital developed algorithms. Remove RapidAI's own models and the platform would still carry partner AI and workflow.

The structure is described in useful detail, and two points of it are left to the site. The cleared triage modules run in parallel with the standard read, with compressed previews for notification only (Rapid LVO, K221248). The FAQ publishes the perfusion thresholds (a relative CBF below 30 percent defines core; only lesions over 3 ml are marked), shows warnings when the contrast bolus or patient motion makes results unreliable, and lets users change thresholds, move the arterial input and reprocess a case.

Navigator Pro lets radiologists review and confirm AI results before they are committed to PACS, but that review is described as optional, and it can send structured AI outputs and measurements straight into the reporting system; EdgeIQ routes positive and negative findings to care teams. Perfusion output is indicated to inform selection for thrombectomy, an invasive decision in a short window. Nothing published covers what happens after the system is wrong, or how a site decides whether AI results may reach PACS or the report without review.

The methods are explained and the models are not. Each cleared module is named with its clinical purpose and its own clearance, and the FAQ explains the perfusion method in close detail: core defined by relative cerebral blood flow below 30 percent, Tmax and MTT distinguished, recommended scan duration, how bolus truncation degrades accuracy and how motion is corrected. The 510(k) summaries describe the detection modules only as AI or neural network software.

No architecture, training data description or model card is published. Upgrades are scheduled with each site's IT team by RapidAI's implementation staff, which is an update practice, but nothing says how a site learns what changed in a model or whether validation carries over.

A short chain for RapidAI's own modules, with the partner and hosting layers not named. The cleared stroke, neurocritical, aneurysm and aortic modules are RapidAI's own, cleared in the name of iSchemaView, Inc., so no outside model provider sits between a scan and those results.

The platform also runs vetted third party algorithms and hospital developed models, and Navigator Pro sells partner algorithms under Rapid names (vertebral compression fracture, cervical spine fracture, incidental pulmonary embolism) without naming who built them. The cloud host behind Rapid Edge Cloud and any subprocessors are not named. An in house, cleared core with the rest of the chain undisclosed.

The headline claim, that the software supports urgent treatment decisions in stroke, rests on prospective, multicenter randomized trials. RAPID perfusion imaging was the patient selection tool in trials that changed stroke guidelines: DEFUSE 3 (stopped early for efficacy of thrombectomy 6 to 16 hours after onset in selected patients) and DAWN (thrombectomy plus medical management superior to medical management alone at 6 to 24 hours), both in The New England Journal of Medicine in 2018, alongside SWIFT PRIME and EXTEND-IA in the earlier window.

Those results extended the treatment window in American Heart Association guidelines to 24 hours. The company's founder led research behind the platform, which is context for independence rather than a bar. The company counts more than 25 trials and 750 publications. Real world figures on its site (a 92 percent reduction in unnecessary transfers, 33 to 52 minutes saved on door to decision or door to needle) come from single hospital case studies with no denominators.

The Rapid LVO 510(k) summary reports standalone sensitivity of 0.96 and specificity of 0.98. The newer modules (aortic, pulmonary embolism, aneurysm tracking) rest on thinner published work than stroke.

Data minimization is described; retention, de-identification and training use are not stated specifically. The Rapid server runs inside the hospital firewall and automatically purges data on a schedule that varies by site. Rapid EdgeIQ processes pixel data and metadata on premises and sends only what each algorithm needs to the cloud, and the 510(k) summaries describe studies being de-identified 'if required' before analysis.

The privacy policy, built around the EU, UK and Swiss Data Privacy Framework, covers personal data generally and mentions sharing with study partners, without saying whether customer imaging is used to develop models. RapidAI configures servers through remote VPN access. General assurances that leave open what is kept, for how long, and whether it reaches model development. Safety engineering, meaning how a missed or wrong result is handled as a safety event, is not published.

HIPAA compliance is claimed and business associate status is not stated. The Enterprise infrastructure page says the Rapid platform is HIPAA and GDPR compliant, and the company is audited for SOC 2 Type 2 and certified to ISO 27701 for privacy information management. Its privacy policy is built around the EU, UK and Swiss Data Privacy Framework and does not address protected health information in the product or the company's role toward US covered entities.

No business associate agreement, tier or subcontractor term is published. The architecture keeps a Rapid server inside the hospital firewall and sends only the data each algorithm needs to the cloud, which narrows exposure but is not stated as a position that the service receives no PHI. Compliance claimed without the status or agreement behind it.

SOC 2 Type 2, ISO 27001 and ISO 27701 are claimed as the company's own; the reports are not public. RapidAI's Enterprise infrastructure page states the platform was developed using the NIST Cybersecurity Framework and is SOC 2 Type 2, ISO 27001, ISO 27701, ISO 42001, ISO 13485 and CSA STAR certified, and HIPAA and GDPR compliant. Servers are configured over site to site or client VPN, and an on-premise fallback keeps AI running if cloud connectivity fails. No trust portal offers the reports or certificates. Recognized attestations named in the company's own material without the artifacts.

The position is stated module by module and each lead clearance can be found. The applicant is iSchemaView, Inc., which trades as RapidAI. Rapid LVO (K221248, 2022; 21 CFR 892.2080, product code QAS) is cleared for triage and notification of suspected ICA or MCA-M1 large vessel occlusion on head CTA, in parallel with the standard read, with previews for notification only.

Other clearances in the company's name include Rapid ICH (K193087), Rapid ASPECTS (K200760), Rapid NCCT Stroke (K222884) and Rapid aneurysm triage (K230074), on top of the original RAPID perfusion software (K121447). The company states its perfusion software holds the only FDA indication for selecting patients for mechanical thrombectomy, a claim about an indication rather than a flag, and in November 2025 announced five further clearances (DeltaFuse, LMVO, midline shift, obstructive hydrocephalus and aortic measurement).

Navigator Pro also offers validated partner algorithms under Rapid names (vertebral compression fracture, cervical spine fracture, incidental pulmonary embolism), whose clearances belong to their developers. Match each module in a quote to its clearance number and indication.

Results compared across patient groups are published for the lead model, and an AI management system is certified. RapidAI's own 510(k) summary for Rapid LVO (K221248) reports sensitivity and specificity with confidence intervals by geography (US and outside the US), by sex, by age group (20 to 39, 40 to 59, 60 and over), by scanner manufacturer (GE, Siemens, Toshiba) and by slice thickness; sensitivity ranged from 0.93 to 1.00 across those groups, with wide intervals in the smaller ones.

The company also holds ISO 42001 certification for its AI management system. The comparison covers the groups the product serves, clinical and technical as well as demographic. It stops short in three places: there are no results by race or ethnicity, the groups are small, and results by group for the other modules are not published. Perfusion analysis is sensitive to acquisition protocol and equipment, and the platform runs in more than 100 countries, yet no module's results are broken down by site type.

An error rate with its method and denominator is published for one module of the neurovascular suite, and nothing else stands behind the output. RapidAI's own 510(k) summary for Rapid Aneurysm Triage and Notification (K230074, cleared July 2023; 21 CFR 892.2080, product code QFM) reports a standalone validation on 266 head CTA cases, 151 positive and 115 negative, against a reference standard set by three experts, in patients aged 21 to 99: sensitivity 0.933 and specificity 0.868, meeting the primary endpoint of an area under the curve of at least 0.95, with results by scanner maker.

That is a test set result for aneurysm triage only. For Rapid LVO (K221248) the summary gives sensitivity of 0.96 and specificity of 0.98 with confidence intervals and subgroup results, but not how many cases were tested. The software shows warnings when bolus or motion problems make results unreliable and lets users change thresholds and reprocess. The company publishes no warranty, indemnity or remediation commitment.

Integration runs through imaging standards into existing systems, with named deployments behind it. The Rapid server is added as a DICOM node between scanners and PACS, results go to PACS, email and the Rapid mobile app, and Navigator Pro connects scanners, PACS, worklists and reporting systems, sending structured AI outputs into the report. The company describes standards based interoperability with PACS, RIS and electronic record systems without naming the record systems or interfaces.

Deployments are named across US health systems including Mayo Clinic, Stanford Health Care, HCA Healthcare, Providence, Intermountain and OSF HealthCare. Standards support with named deployments behind it.

Hybrid on-premise and cloud deployment is documented; hosting regions are not. A Rapid server, physical or virtual, sits inside the hospital firewall as a DICOM node, and Rapid Edge Cloud adds cloud processing with an on-premise fallback if connectivity fails. Rapid EdgeIQ processes pixel data and metadata on premises and sends only what each algorithm requires to the cloud.

The server purges data automatically on a site specific schedule, and RapidAI configures it through remote VPN access. The company names no cloud provider or regions, does not say whether processing crosses a border for the platform's customers in more than 100 countries, and says nothing about tenant isolation.

No price is published, and how the product is bought is discoverable. RapidAI sells the Rapid Enterprise Platform with separately cleared modules across stroke, neurocritical care, aneurysm, aortic and pulmonary embolism imaging, partner algorithms available through the same platform, and services: implementation that typically takes a site live within two weeks, a server inside the firewall, 24/7 support, and education through RapidU. The route in is a demo.

What is not published is the unit of charge: whether a customer pays for the platform, for modules one by one, per site or per study, and how the price changes as new modules are cleared. No price, and the shape of the offer is discoverable.

Deep in stroke and neurocritical imaging and widening from there, with validation behind the core. The platform covers ischemic stroke from non-contrast CT through CTA, CT and MR perfusion and intervention; hemorrhage, subdural hematoma, midline shift and hydrocephalus; aneurysm detection and tracking; aortic measurement and pulmonary embolism; and, through partners, spine and other fractures, lung nodules, mammography and more.

It reaches radiologists, stroke and neurointerventional teams and care coordinators in more than 2,500 hospitals and supports clinical trials. Stroke selection is validated in randomized trials; the newer and partner modules are named with less published validation by setting. Named coverage with validation behind part of it.