Company Spotlight: Qure.ai qER-CTA - AI Stroke Triage and the 6.36-Minute Notification Race

Sam: Welcome to AI in Medicine - Smart Summaries. Today, we are looking at a triage tool that identifies suspected large vessel occlusions, with a tested mean notification time of exactly 6.36 minutes.

Maya: We are breaking down the official FDA 510(k) summary for qER-CTA (v1.0) by Qure.ai Technologies, which was published by the U.S. Food and Drug Administration on September 8, 2025.

Sam: And just a quick reminder before we dive in, our voices are AI-generated, and this episode is a summary of a public FDA document, not company marketing material.

Maya: Exactly. This is a Company Spotlight episode, meaning we are looking closely at one specific product clearance to understand what it does, how it was tested, and what it changes on the ward or in the boardroom.

Sam: So let us set the stage. The document we are looking at clears a piece of software called qER-CTA. The FDA classifies it as Radiological Computer Aided Triage And Notification Software. It is a Class II device.

Maya: Right. And the specific clinical target here is something called a Large Vessel Occlusion, or LVO. The indications for use state that the device analyzes CT angiogram images of the brain acquired in the acute setting.

Sam: An acute setting. Meaning an emergency room or a stroke center where time is obviously critical. What exactly is the software doing when it looks at those images?

Maya: According to the summary, it uses a deep learning algorithm to analyze the internal carotid artery, which is abbreviated as ICA, and the M1 segment of the middle cerebral artery, or MCA. It is looking for LVOs on scans of adults, specifically defined as patients 22 years of age or older.

Sam: Okay, so it is looking at specific arteries in the brain for a blockage in adults. If it finds one, what happens next? Does it just pop up on the screen and say, hey, look at this?

Maya: This is where the wording of the 510(k) summary is extremely specific and important for hospital buyers to understand. It is described as a notification-only, parallel workflow tool. It works independent of the standard of care workflow.

Sam: Parallel workflow. Meaning the normal process of the scan going to the radiologist happens exactly as it always does, and this software runs off to the side at the same time?

Maya: Exactly. The document says it notifies an appropriate medical specialist in parallel to standard of care image interpretation. But the summary is very clear about its limitations. It says the identification of suspected findings is not for diagnostic use beyond notification.

Sam: So the AI is essentially tapping a specialist on the shoulder saying, you might want to look at this, but it is not formally diagnosing the condition itself.

Maya: That is right. In fact, under the device description, it explicitly states that qER-CTA is not intended to direct attention to specific portions of the image, rule out target conditions, or be used as a standalone tool for clinical decision-making.

Sam: I want to get into the weeds on how that shoulder tap actually happens, because the document outlines the workflow, and it sounds a bit different than what people might expect from an AI alert.

Maya: It does. The summary says it provides a case-level output available in the PACS or workstation for worklist prioritization or triage. But here is the kicker. It says, and I quote, it does not alter the original image, change the worklist order, or send proactive alerts directly to the end user.

Sam: Wait, really? It does not change the worklist order and does not send proactive alerts directly to the user?

Maya: That is exactly what it says. Instead, it says the end user can sort the worklist based on the passive notification flag.

Sam: That is a massive workflow detail for a health-system leader to consider. If the tool relies on a passive notification flag, that means the clinician has to actively choose to sort their worklist by that flag to see who the AI thinks has a large vessel occlusion.

Maya: Yes, from a human factors perspective, that is a significant distinction. It is not interrupting them; it is waiting for them to look for it. Though the document also notes that images can be previewed through a mobile application.

Sam: Right, the mobile app. That seems to be a big selling point for these triage tools, getting the scan to the specialist wherever they are. But the FDA summary has some strong warnings about that mobile app, doesn't it?

Maya: It has very strict guardrails. It states that images previewed through the mobile application are compressed and for informational purposes only. It specifically says they are not intended for diagnostic use beyond notification.

Sam: So a neurosurgeon or interventional neurologist might get the notification on their phone, see the compressed image, and realize they need to act, but they cannot legally or clinically base their final decision on that phone screen.

Maya: Correct. The document says notified clinicians are responsible for viewing non-compressed images on a diagnostic viewer, conducting appropriate patient evaluation, and engaging in relevant discussions with the treating physician before making care-related decisions or requests.

Sam: That covers the what and the how. Let us talk about the regulatory pathway, because that tells us a lot about the competitive landscape. This is a 510(k) clearance, which means Qure.ai had to prove this device is substantially equivalent to something already on the market.

Maya: Yes, they used a predicate device. The primary predicate device listed here is Viz LVO, manufactured by Viz.ai, Inc., under the 510(k) number K223042.

Sam: Viz.ai is obviously a major player in this specific space. The document actually provides a head-to-head comparison table between Viz LVO and qER-CTA. What stands out to you in that comparison?

Maya: A few things. First, the intended users. The predicate device, Viz LVO, lists its intended users as neurovascular specialists, such as vascular neurologists, neuro-interventional specialists, or users with similar training.

Sam: And how does qER-CTA define its users?

Maya: It goes a bit wider. It lists neurologists, neuroradiologists, radiologists, neuro-interventional specialist, and/or other emergency department physicians.

Sam: Including emergency department physicians. That is interesting from a commercial standpoint. It suggests they are positioning this not just for the stroke specialists at the hub hospital, but for the ED docs at the spoke hospitals who are doing the initial scanning.

Maya: That aligns with the acute setting indication. Now, let us look at the performance testing, because that is where the real numbers are. They conducted a standalone study to evaluate qER-CTA.

Sam: A standalone study. Meaning this was likely a retrospective analysis of scans, rather than a prospective clinical trial where they tracked patients live in the ER. The summary does not mention a prospective trial.

Maya: That is right. The summary says the target sample size for the study was 584 head CTA scans. And the split was very close. It included 289 LVO cases and 295 non-LVO cases.

Sam: I will point out that having 289 positive cases and 295 negative cases does not reflect the real-world prevalence of large vessel occlusions in an ER. The dataset was clearly enriched with positive cases to rigorously test the algorithm's ability to find them.

Maya: Exactly. To establish the ground truth for these 584 scans, they used three U.S. board-certified neuroradiologists, each with at least 10 years of experience.

Sam: That is a solid reference standard. So, how did the AI perform against those three veteran neuroradiologists?

Maya: The subject device, qER-CTA, achieved a sensitivity of 91.35%, with a 95% confidence interval of 87.54% to 94.07%.

Sam: Sensitivity of 91.35%. That is the software's ability to correctly identify the large vessel occlusion when it is actually there.

Maya: Right. And for specificity, it achieved 91.86%, with a confidence interval of 88.18% to 94.47%.

Sam: Specificity of 91.86%. So when there is no LVO, it is highly accurate at identifying the scan as negative, which helps keep false positives relatively low.

Maya: The Area Under the Curve, or AUC, was 0.959.

Sam: For comparison, because the document puts them side-by-side, what were the numbers for the predicate device, Viz LVO?

Maya: The table lists Viz LVO's sensitivity at 87.8% and its specificity at 89.6%, with an AUC of 0.91.

Sam: So qER-CTA's numbers are mathematically a bit higher in this specific table. But we have to be very careful here as an independent podcast. A 510(k) clearance does not mean the FDA has declared the new device clinically superior.

Maya: That is a crucial point. The conclusion section of the summary simply states that the qER-CTA device is substantially equivalent to the predicate device. It says the new device does not introduce fundamentally new scientific technology.

Sam: Right. They are in the same regulatory bucket. Now, what about the speed? You mentioned in the hook that it was fast. What exactly did the study measure?

Maya: They measured Time to Notification. For the 584 cases, the mean time to notification of a specialist for LVO cases was 6.36 minutes. The standard deviation was 3.68, and the median time was 6.89 minutes.

Sam: A mean of 6.36 minutes. And how did that compare to the predicate in the table?

Maya: The table lists the mean time to notification for Viz LVO as 7.32 minutes.

Sam: So both are operating in that six to seven-minute window. In an acute setting, where a radiologist might have a backlog of scans on their standard of care worklist, getting a notification in six and a half minutes is why these triage tools are being heavily adopted.

Maya: Absolutely. But remember the workflow limitation we discussed earlier. It takes a mean of 6.36 minutes to generate that passive notification flag. The clinician still has to see the flag and act on it.

Sam: That is such an important distinction. The AI might do its job in 6.36 minutes, but if the doctor is not sorting their list by that flag, or if they are waiting to review it on a diagnostic viewer, the actual clinical time could be longer.

Maya: Speaking of getting the software to talk to the hospital's systems, the device description outlines two alternatives for how users can engage with qER-CTA from an IT perspective.

Sam: What are the options? Because IT integration is often the biggest hurdle for deploying AI in a hospital network.

Maya: The first alternative is for de-identified CTA scans to be sent to qER-CTA via transmission functions built within the user's PACS or workstation. The results are then pushed back once processing is complete.

Sam: So it integrates directly with the Picture Archiving and Communication System, the PACS. That is standard. What is the second alternative?

Maya: The second alternative is for client systems that do not have de-identification and re-identification capabilities. In that case, qER-CTA interacts with an on-premises gateway rather than directly with the PACS.

Sam: That is a smart technical workaround. If a hospital's PACS cannot strip out patient health information before sending the data out for AI analysis, this on-premises gateway handles the privacy side of things locally before anything leaves the network.

Maya: Exactly. The software verification and validation testing followed FDA guidelines, and the summary notes they used standards like ISO 13485:2016 and IEC 62304:2006 plus A1:2015 for the medical device software lifecycle.

Sam: So they ticked all the necessary software engineering boxes. Before we wrap up, let us summarize the most important limitations stated in the document, because any clinician using this needs to know what it cannot do.

Maya: First, qER-CTA should not be used in lieu of a full patient evaluation or relied upon to make or confirm a diagnosis. Second, it is limited to analyzing the ICA and the M1 segment of the MCA for LVOs. Third, it is only for adults 22 years of age or older.

Sam: And fourth, it does not alter the original image or proactively push alerts directly to the user; it uses a passive notification flag that you have to look for.

Maya: And finally, any images previewed on the mobile application are compressed and for informational purposes only.

Sam: Those are clear, rigid boundaries set by the FDA. It is a powerful tool to run in the background and flag suspected LVOs with over 91.35% sensitivity, but the final diagnostic burden remains entirely on the clinician at the workstation.

Maya: To recap, today we reviewed the 510(k) summary for Qure.ai's qER-CTA, cleared on September 8, 2025. It demonstrated substantial equivalence to the Viz LVO predicate device in analyzing head CTA scans for suspected large vessel occlusions, achieving a mean time to notification of 6.36 minutes.

Sam: As always, you can find a link to the original FDA summary document in our show notes. Read it for yourself, especially if you are evaluating triage software for your facility.

Maya: And remember, we are AI voices summarizing a public regulatory document. This podcast is for informational purposes only and is not medical advice.