Company Spotlight: Cercare Medical Virtual Expert - Mapping Brain Tumors and Blood Flow

Sam: What if the key to understanding a brain tumor wasn't just measuring how wide it is, but mapping exactly how the blood flows through its microscopic vessels, using an AI tool branded as a Virtual Expert?

Maya: Today we are breaking down the official U.S. Food and Drug Administration 510(k) summary for the CMN Capillary Function with Oncology Virtual Expert, published by the FDA on June 22, 2026, for the manufacturer Cercare Medical A/S.

Sam: As a quick reminder, our voices are AI-generated, and this episode is a summary of a publicly available document, not an advertisement or company-sponsored material.

Maya: This is a Company Spotlight episode, meaning we are looking closely at one specific product clearance to see what it actually does, what the testing data really says, and what it means for the clinicians who might use it and the companies competing in this space.

Sam: And this document is fascinating because it tackles a really complex area of neuro-oncology. We are talking about brain cancer. Specifically, the document says this is for a subset of primary brain tumors, brain glioma, WHO Grade II through IV.

Maya: That is exactly the patient population. The indications for use state this is for adult patients who are known to have pathologically proven glioma. It specifically lists glioblastoma, which is WHO Grade IV, astrocytoma, which spans WHO Grade II, III, and IV, and oligodendroglioma, which is WHO Grade III.

Sam: Right, so a patient already has a proven diagnosis. This tool is not finding the cancer for the first time. The document explicitly says, Oncology Virtual Expert is not to be used for primary diagnosis.

Maya: Exactly. It is intended to be used by trained professionals as an additional source of information for decision support. But what it actually does as decision support is quite comprehensive. It provides semi-automatic labeling, visualization, and volumetric quantification of these gliomas.

Sam: And the way it does that is by plugging into a larger platform. The document describes the Cercare Medical Neurosuite, or CMN for short. CMN is the base image processing software. Then they have these modules you can add on.

Maya: Yes, it is a modular software-only device. The base platform handles image viewing and registration of DICOM images from MRI and CT scanners. Then you have the Capillary Function module, which is identical to an older reference device they cleared. And now, they are adding the Oncology Virtual Expert module, which is the subject of this new clearance.

Sam: Let us talk about that Capillary Function module first, because the document goes into some detail on it. It says it provides analysis capabilities for functional and dynamic imaging datasets acquired with MRI. What does that actually mean in plain English?

Maya: It means it calculates parameters related to tissue flow, or perfusion, and tissue blood volume. It is used for visualization and analysis of dynamic imaging data, showing properties of changes in contrast over time.

Sam: The document specifically notes it uses a Dynamic Analysis Module for dynamic contrast-enhanced imaging data for both MRI and CT. That sounds like highly specialized physiological mapping.

Maya: It is. And in addition to that dynamic contrast-enhanced data, the Capillary Function module also has a Diffusion Weighted MRI module to visualize local water diffusion properties. That is the physiological baseline the system is operating on.

Sam: Okay, so then we add the Oncology Virtual Expert module. The summary says it provides delineation and volumetric quantification of perceived pathological tissue based on primary acquired MRI data. Specifically, T1 images acquired pre- and post-contrast, and T2FLAIR images.

Maya: And here is where it gets interesting for clinical workflows. It is not just drawing a circle around a tumor. It separates the tissue into classifications. For the US market, it delineates T2/FLAIR hyperintensity and T1 enhancing tissue.

Sam: Wait, you said for the US market. Does it do something different elsewhere?

Maya: Yes, it does. The document explicitly notes a difference in features. In non-US markets, the software delineates T2/FLAIR hyperintensity, T1 enhancing tissue, and necrotic tissue. But in the US market, necrotic tissue is excluded from that list.

Sam: That is a fascinating regulatory detail. The FDA document does not explain why necrotic tissue was left off the US feature list, it just states that it is.

Maya: Correct. We do not have the medical or regulatory reasoning for that omission in this summary. But what we do have is how the Oncology Virtual Expert interacts with the Capillary Function module. And this is the core value proposition described in the document.

Sam: Right, the document says it allows for the automatic extraction of perfusion data from each delineated lesion type. So, it is automatically pulling that tissue blood volume and flow data specifically from the T1 enhancing area or the T2/FLAIR area.

Maya: Exactly. You get the mean value and standard deviation for the perfusion images within those specific segmented tumor lesions. It merges the structural volume of the tumor with the physiological blood flow metrics inside it.

Sam: That definitely answers the question of what this changes on the ward. But let us look at the boardroom side. This is a 510(k) clearance, which means Cercare had to prove this device is substantially equivalent to something already on the market.

Maya: They did. They used a predicate device named NS-HGlio, manufactured by Neosoma, Inc. The FDA cleared that predicate device under the 510(k) number K221738.

Sam: And when you look at the substantial equivalence comparison table in the document, they share the same product codes, QIH and LLZ, which fall under the regulation name Medical Image Management and Processing System.

Maya: Both the subject device and the predicate device are indicated for semi-automatic labeling, visualization, and volumetric quantification of gliomas. They both use T1 pre-contrast, T1 post-contrast, T2, and FLAIR MRI images.

Sam: But there are two major differences noted in the summary. The first one is how the software actually runs. The Neosoma predicate relies on off-premises resources. It is a cloud-based platform only.

Maya: Whereas the Cercare device does not rely on off-premises resources. It runs on a standard off-the-shelf PC workstation or a virtual platform such as VMware. The document concludes this difference does not impact the intended use or raise new questions with the safety and performance of the device.

Sam: That is a huge point for hospital IT buyers. Some health systems heavily prefer local or virtual machine deployments behind their own firewalls rather than sending massive DICOM studies to a third-party cloud. The FDA clearing this locally is a distinct infrastructural choice.

Maya: The second major difference is the perfusion extraction. The document states that the computation of perfusion metrics within the different tissue classes is a unique data collection and reporting feature for the Subject Device. The predicate device did not have that.

Sam: So how did Cercare prove to the FDA that their tool actually works accurately? Let us get into the real numbers from the performance data section.

Maya: They ran comparative performance testing using a 100-patient subset of the MU-Glioma Post dataset from the University of Missouri. This subset represented 425 studies.

Sam: And the breakdown of those 100 patients is highly specific. They included glioblastoma with n=73, astrocytoma with n=16, diffuse glioma with n=8, oligodendroglioma with n=2, and pilocytic astrocytoma with n=1.

Maya: To establish what was accurate, both the Cercare device and the Neosoma predicate device were independently compared against the same expert manual segmentation ground truth.

Sam: And how did they determine that ground truth? Because having two experts look at a brain scan can sometimes result in two different maps of a tumor.

Maya: The summary details a rigorous process. It was established through a STAPLE-based consensus algorithm with three-tiered expert review, including final consensus approval by two board-certified neuroradiologists.

Sam: Okay, so they had a very solid benchmark. What were the actual performance results?

Maya: The metric used is the mean Dice similarity coefficient. For the delineation and volumetric analysis of the T1 enhancing region, the Subject Device achieved a mean Dice similarity coefficient of 0.706.

Sam: And for the T2/FLAIR hyperintensity region?

Maya: For that region, the mean Dice similarity coefficient was 0.822. The document notes that for both of these metrics, the Predicate Device was comparable.

Sam: Let us pause on those numbers for a second. 0.706 and 0.822. For anyone building AI in this space, these numbers show exactly what the bar is for substantial equivalence right now. It is 0.706 for T1 enhancing and 0.822 for T2/FLAIR.

Maya: That is an excellent point. And the document includes an important contextual note about those scores. It says performance increased with lesion volume, as expected from the properties of the Dice similarity metric. So larger tumors tended to score higher on similarity than smaller ones.

Sam: The testing also covered different hardware. It mentions the dataset was acquired predominantly on Siemens MRI systems at 1.5T and 3.0T field strengths. But it also notes that subgroup analyses demonstrated consistent performance across scanner manufacturers, listing Siemens, GE, and Philips.

Maya: Yes, but there is a major caveat regarding scanners that clinicians need to know about. The system can collect and report volumetric data across time if more studies exist for the same patient. This is crucial for tracking disease progression.

Sam: Right, longitudinal comparison. What is the catch?

Maya: The indications for use include a very specific warning: to minimize measurement variability, such longitudinal comparisons should be performed using the same scanner model and consistent acquisition parameters.

Sam: Wow. The exact same scanner model. That is a massive practical limitation. If a patient gets their baseline MRI on a Siemens 1.5T and a later follow-up on a GE 3.0T, the labeling implies you should not rely on the software's automatic volume comparison.

Maya: That is exactly what the labeling instructs. The variability between different hardware models and scanning protocols could compromise the volumetric comparison. This is a perfect example of why reading the actual FDA summary is so important, rather than just looking at a marketing sheet.

Sam: The document also makes sure to bracket the overall clinical responsibility. It says, CMN is intended to be used as decision support software only and the clinician continues to provide all treatment decisions.

Maya: It goes even further than that in the Technological Characteristics section. It states, it is emphasized that the output from the software should not be used without close scrutiny of the relevance to the patient in question, including all available evidence and quality checking of the appropriateness of the output.

Sam: That is a very strong disclaimer. It is basically saying, the AI might give you a volume and a perfusion map, but you better check its work.

Maya: It is standard for Class II medical image processing systems, but it is explicitly spelled out here. The device is not a dedicated PACS system for long term persistent storage of patient data either. It processes the DICOM images and generates the reports, which can then be sent to the hospital's local PACS via network communication.

Sam: Speaking of regulations, the document lists the performance standards Cercare conformed to. It is a long list of ISO and IEC standards that medtech developers will recognize. ISO 13485 for quality management, ISO 14971 for risk management, IEC 62304 for software lifecycle processes.

Maya: They also followed the FDA guidance titled, Computer-Assisted Detection Devices Applied to Radiology Images and Radiology Device Data – Premarket Notification 510(k) Submissions, from September 2022.

Sam: The FDA itself also includes some boilerplate warnings in the letter to the company. They note that CDRH does not evaluate information related to contract liability warranties. And they remind the company that device labeling must be truthful and not misleading.

Maya: Which is standard language in a 510(k) clearance letter, but always worth remembering when we evaluate AI products. The FDA is clearing the device based on substantial equivalence to the predicate, in this case the Neosoma NS-HGlio, and determining it presents no unacceptable risks to the intended patient population.

Sam: So, looking at the big picture of this Company Spotlight, Cercare Medical managed to take a base platform that already did capillary function and diffusion mapping, and they legally merged it with an Oncology Virtual Expert module to automatically extract perfusion data from specific tumor zones, all running locally instead of in the cloud.

Maya: And they proved it works well enough to be cleared by testing it on a retrospective dataset of 425 studies, achieving Dice similarity scores of 0.706 and 0.822 against a strict board-certified neuroradiologist ground truth.

Sam: Maya, what is the one memorable takeaway from this document for our listeners?

Maya: The most striking takeaway is the strict instruction regarding longitudinal tracking. The software can automatically compare tumor volumes over time, but only if you use the exact same scanner model and consistent acquisition parameters. It highlights how sensitive these AI tools are to the physical hardware capturing the data.

Sam: That is a brilliant point. The Virtual Expert is only as consistent as the MRI machine feeding it.

Maya: That wraps up our Company Spotlight on Cercare Medical's CMN Capillary Function with Oncology Virtual Expert. We unpacked the indications for use in brain glioma, the technical architecture, and the real numbers behind its clearance.

Sam: You can find a link to the full FDA 510(k) summary in the show notes. As always, this podcast is for informational purposes only and is not medical advice. Thanks for listening to AI in Medicine - Smart Summaries.