RapidAI

The algorithms are the product. Imaging arrives from scanners the hospital already owns, the models analyse it, and what is sold is the analysis. There is no hardware, no reading service and no record system underneath.

The module structure reinforces rather than dilutes this: each cleared capability is a distinct model addressing a distinct clinical question, from vessel occlusion detection to midline shift quantification, assembled on a common platform. What the platform provides is delivery and integration; what the customer buys is the set of models.

Adjunctive by design, with a physician making every decision, and the interesting part is how much further than alerting the output goes. The company describes its approach as delivering not just triage notification but visualisation, localisation, characterisation and tracking of change over time, which gives the clinician material to reason with rather than a flag to accept or dismiss.

The consequence of the output is unusually high even so. Perfusion analysis used to select patients for mechanical thrombectomy feeds directly into whether a patient undergoes an invasive procedure inside a narrow time window, and the company holds an indication specifically for that selection. Held at B because nothing published describes what a clinician is shown when the model is uncertain, or how disagreement between the software and the reading physician is expected to resolve.

Scope is documented module by module rather than in aggregate, which is the useful form for a multi indication platform. Each cleared module is named, its clinical purpose stated and its clearance reported separately, so a buyer can tell exactly what is authorised rather than inferring it from a platform claim.

Some performance detail is published alongside, including a reported reduction of over 36 percent in radiologist comparison time for serial scan alignment. No model architecture or training data description was located, and the aggregate claim of more than 700 clinical studies is a count rather than a synthesis, so a reader has to go to the literature to know what those studies found.

Nothing identifies any party in the chain: no model or model family, no hosting arrangement and no sub processor list was located in two passes, and no retention schedule or encryption detail was found. One property of the workload is worth naming because it establishes that the answers exist somewhere. Stroke imaging is acute, so studies move through the system within minutes of acquisition, and deployment spans more than a hundred countries.

A company operating at that tempo across that many jurisdictions has necessarily resolved where processing happens, which transfers are lawful and what residency applies in a great many places, because it could not operate otherwise. Those answers are simply not public, which is a different situation from a vendor that has not thought about it, and it means a buyer asking is requesting disclosure rather than prompting a decision.

The acuity also shapes the retention question in a specific way. An emergency workflow has an obvious operational reason to hold a study briefly and no obvious reason to hold it afterwards, so a short stated retention period would be easy to give and would distinguish this vendor. Ask for the residency and transfer position for your jurisdiction, retention after a case closes, a sub processor list, and whether processed studies contribute to model development.

Probably the deepest evidence base in the imaging lane of this index, and it rests on something rarer than volume.

Research using this perfusion software contributed to the trials that extended the stroke treatment window in national guidelines. That is a different order of claim from a vendor demonstrating its product works: the software was the selection instrument in studies that changed what treatment patients are eligible for, so the evidence is embedded in clinical practice rather than sitting beside it. The company also holds an indication specifically for selecting patients for thrombectomy, which required its own evidence.

Around that sit more than 700 stated clinical studies and 28 abstracts at a single international stroke meeting in 2026. Deployment across 2,500 hospitals in over 100 countries corroborates adoption without being counted as evidence in itself.

Graded on an honest basis. No published stewardship position, retention schedule or encryption detail was located in this pass.

The workload has a property worth noting: stroke imaging is acute, so images move through the system within minutes of acquisition and across more than 100 countries, which means the residency and transfer questions are answered somewhere for a great many jurisdictions. Those answers are not public.

Graded on an honest basis and flagged for re verification. No compliance statement or agreement posture was located in this pass.

Operating in over 100 countries means many regimes apply simultaneously, and a national partnership covering 20 health clusters in Saudi Arabia sits under a framework different again from either United States or European law. Nothing published sets out how those are handled.

Recorded honestly: the dedicated trust and security search this index requires was not run in this pass, so the grade is provisional and should not be quoted until it has been. No attestation was encountered incidentally.

What is known points upward and should temper any inference from the absence. A platform embedded in 2,500 hospitals across 100 countries, running inside image archiving and record infrastructure, has passed a very large number of institutional security reviews.

Deep and still widening. Five further modules were cleared in November 2025 covering serial scan alignment, vessel occlusion on angiography, midline shift quantification, suspected obstructive hydrocephalus and aortic measurement, each addressing a separate clinical question and each cleared on its own.

The distinguishing authorisation is narrower and more consequential than the count. The company states it holds the only perfusion software indicated for selecting patients for mechanical thrombectomy. That is an indication to inform who receives an invasive procedure, not merely to flag a study for attention, and it is a materially harder thing to obtain than a triage clearance. The claim of uniqueness is the vendor's own and is checkable against the clearance database.

No subgroup performance, monitoring policy or bias evaluation was located in company material, and the grade is C rather than lower because a literature base of this size very likely contains demographic analysis that was not retrieved in this pass. Re verify against the published studies.

The concern specific to this product is geographic rather than only demographic. Deployment spans more than 100 countries with very different imaging equipment, protocols and patient populations, and perfusion analysis in particular is sensitive to acquisition parameters. A model validated principally in North American and European comprehensive stroke centres is being applied in settings that differ on all three. Nothing published addresses how performance holds across them.

Scope is documented module by module rather than in aggregate, which is the right form for a multi indication platform and directly answers the failing this index recorded against a competitor in the same cluster. Each cleared module is named, its clinical purpose stated and its clearance reported separately, so a buyer can establish exactly what is authorised for the module they are deploying rather than inferring authorisation from a platform level claim.

That distinction matters because a platform accumulates modules over years and a customer buying the newest one is otherwise relying on the credibility of the oldest. Some performance detail is published alongside, including a reported reduction of more than a third in radiologist comparison time for serial scan alignment, which is an operationally meaningful figure for a workflow product. Held at C because the evidence is a count rather than a synthesis.

More than seven hundred clinical studies is a quantity, not a finding, and a reader has to go to the literature themselves to learn what those studies concluded or whether any of them disagreed. No model architecture, training data description or per module sensitivity and specificity was located, and no warranty, indemnity or remediation commitment. Ask for per module performance on the indications you are deploying, and for the three studies the company considers most probative rather than the count.

Integration is treated as a product property rather than a project. Modules are delivered through a cloud platform and a radiology application described as operating across image archiving and record systems so that findings reach the radiologist inside the existing workflow rather than in a separate viewer.

Stroke imposes the hardest version of this requirement, because the value of the analysis decays in minutes and any step that asks a clinician to leave their workflow destroys it. Operating at this scale is reasonable evidence the integration works. Held at B because no interface standard, named record vendor certification or write back mechanism was located.

The platform is delivered through what the company calls an edge cloud, which implies processing close to the hospital rather than purely in a central service, and that architecture is the right answer to the constraint this product faces: a stroke study cannot wait on a long round trip.

Operating across more than 100 countries means residency has been solved repeatedly, including inside a national health system arrangement in the Middle East. Held at B because the specific split between edge and cloud processing, the regional hosting arrangements and the retention schedule for images are not published.

Nothing published. No price, no pricing mechanism and no unit of sale, and the modular structure makes the omission more consequential than usual because a buyer cannot tell whether they are purchasing a platform, a bundle or individual cleared modules, or how the price changes as modules are added.

That matters commercially as well as practically. A platform that grows by adding separately cleared modules has an obvious expansion path, and a hospital committing to the platform without knowing the cost of the modules it will want later is accepting an open ended position.

Deep in one clinical domain and widening from it. The centre of gravity is neurovascular and specifically stroke, where the platform covers detection, perfusion analysis, treatment selection and change over time, and recent clearances extend into traumatic brain injury, hydrocephalus and aortic measurement.

The setting is the acute hospital and the emergency imaging pathway, reaching stroke teams, neurointerventionalists and radiology departments. Geographic reach is the broadest of any imaging record here, at more than 2,500 hospitals in over 100 countries, including a national scale arrangement in the Middle East. Nothing addresses ambulatory or outpatient imaging.