Company Spotlight: iSchemaView Rapid Aortic Measurements - The Shift from Neuro to Aorta

Sam: When a new medical AI claims 100% agreement with clinical experts across numerous vessel segments, you have to ask, what exactly is it measuring, and what are the blind spots? Today, we are looking at the FDA 510(k) summary for Rapid Aortic Measurements, published by the U.S. Food and Drug Administration on September 23, 2025, for iSchemaView, Inc.

Maya: And before we jump in, a quick reminder that our voices are AI-generated. This episode is a summary of a publicly available FDA document, not company marketing material, and as always, nothing we discuss is medical advice.

Sam: Welcome to another Company Spotlight. We are looking strictly at what the FDA cleared, how the testing was done, and what it actually means for hospitals. Maya, set the stage for us. What exactly is Rapid Aortic Measurements?

Maya: It is an image analysis and measurement device, specifically classified as a Medical Image Management And Processing System. It is designed to evaluate the aortic and iliac arteries using CT imaging of the chest, abdomen, and pelvis. The software can process both contrast-enhanced and non-contrast CT datasets.

Sam: So a patient comes into the emergency department, gets a CT scan of their torso, and this software sits in the background analyzing the blood vessels?

Maya: Exactly. The document says the module segments the aorta, the iliac arteries, and major branching vessels. It essentially identifies the boundaries of these vessels and removes obstructive bones and other tissues from the image to give doctors a clear view. It then provides both 2D and 3D visualizations of those segmented vessels.

Sam: What kind of outputs are the doctors actually looking at on their screens?

Maya: The summary lists specific outputs. Doctors get centerline measurements of the aorta and iliacs, fixed measurements, and something called Aortic Zone Measurements, which specifically look at the Maximum Oblique Diameter. It also generates 3D Volume Renderings, Rotations, Curved Planar Reformations of the isolated left and right iliacs, and aortic oblique Multiplanar Reconstructions.

Sam: That is a lot of imaging jargon, but basically, it reconstructs the raw CT slices into 3D models and flattened, curved views so a vascular surgeon or radiologist can instantly see the exact diameter of the aorta at various points.

Maya: Yes. And crucially, it includes Longitudinal Tracking visualizations. That means if a patient has a known issue and they come in for their next scan, the software helps track changes over time by comparing it to previous scans.

Sam: Which is exactly what you want when deciding if a patient needs an intervention today or can wait. Who is the intended user for this tool?

Maya: The document explicitly lists radiologists, cardiothoracic surgeons, vascular surgeons, endovascular specialists, ED physicians, or users with similar training. And it notes that Rapid Aortic Measurements is indicated for adults.

Sam: Let us talk about how this got cleared. This is a 510(k) summary, which means iSchemaView did not have to prove this was a brand new invention, right? They just had to prove it was substantially equivalent to something already on the market.

Maya: That is right. The 510(k) pathway requires a predicate device. Interestingly, the predicate device claimed here is Rapid Neuro3D, which is also marketed as Lumina3D. That device was cleared under a previous submission, K243350.

Sam: Wait, Neuro3D? So the predicate device was for the brain, and this new clearance is for the chest and abdomen? That is a major shift.

Maya: Exactly. The FDA summary includes a side-by-side comparison table. Rapid Neuro3D is an image analysis software for conventional CT Angiography from the aortic arch to the vertex of the head. It provides visualization of the neurovasculature supplying arterial blood to the brain. Rapid Aortic Measurements uses the same technological characteristics, traditional AI and ML, but applies it lower down in the body to the chest, abdomen, and pelvis.

Sam: The company is taking the engine they built to look for brain vessel issues and retraining it to look at the massive vessels in the torso. Does the document say if there are differences in how both systems handle the data?

Maya: There are some notable differences in the technical constraints, likely because scanning a torso is different from scanning a head. For instance, the neuro device excluded slice thicknesses greater than 1.5mm. The new aortic device is a bit more forgiving; it excludes slice thicknesses greater than 3 mm.

Sam: That makes sense. Body CT slices are often thicker than high-resolution head CT slices. Let us get into the performance data. How did they actually prove this thing works, and what are the numbers?

Maya: The company conducted performance validation testing using a dataset that was independent from the data used during model training. The document states the study design used 108 cases. These included non-contrast CT, CT Angiography, or contrast-enhanced CT scans.

Sam: 108 cases. Is there any breakdown of who those patients were?

Maya: There is, but the numbers in the summary are a little confusing. The document says the study used 108 cases from 115 patients. It breaks down the locations as 54 US, 24 OUS, which usually means outside the US, and 30 unknown.

Sam: Wait, the document lists 54 US, 24 OUS, and 30 unknown. But how do you get 108 cases from 115 patients? Usually, you have more cases than patients if some people get scanned several times, not fewer.

Maya: The document does not explain that discrepancy. It also breaks down the demographics, stating there were 70 male and 38 female patients, aged from 22 to 90+.

Sam: And there is another mathematical quirk. The breakdown of 70 male and 38 female patients does not seem to align perfectly with the 115 patients total mentioned right before it. Does the FDA summary elaborate on this missing information at all?

Maya: It does not. It simply presents those numbers exactly as I read them. It also mentions that the data included representation from all major CT scanner manufacturers, specifically naming Siemens, GE, Toshiba/Canon, and Philips.

Sam: Okay, so there is some sloppy math in the summary text regarding the patient count, but let us look at the actual testing. What did they measure in those 108 cases?

Maya: They assessed numerous vessel segments. Specifically, 2,268 vessel segments for the 3D volume rendering images, 756 segments each for the stretched and straightened curved planar reformations, 324 for oblique multiplanar reconstructions, and 666 for measurement outputs.

Sam: How did the AI perform against human experts?

Maya: The results reported are flawless. For the primary endpoint of segmentation quality and clinical accuracy, the document states it was determined by the consensus of up to 3 clinical experts against the source DICOM images. It passed with 100% agreement.

Sam: 100% agreement. That is a bold number. I do notice the phrasing up to 3 clinical experts. That implies some scans might have been reviewed by all of them, but others might have only been reviewed by fewer experts.

Maya: It does not specify. It just says up to 3. But the 100% agreement held across the board. The CPR and MPR quality passed with 100% agreement. The anatomical labeling passed with 100% agreement. And the maximum oblique diameter location accuracy passed with 100% agreement between readers for all segments reviewed.

Sam: Those are qualitative endpoints, right? Essentially doctors looking at it and saying, yes, this looks correct. Did they do any hard mathematical measurements to see how far off the AI boundaries were from a human boundary?

Maya: Yes, they did a segmentation accuracy study to quantitatively assess the outputs against the ground truth. For the volume rendering outputs, the primary endpoint passed with an average Dice coefficient of 0.93 and an average Hausdorff distance of 0.54 mm.

Sam: Let us translate those for the audience. The Dice coefficient measures overlap, so 1.0 would be a perfect pixel-for-pixel match between the AI and the human. So 0.93 is very strong overlap. And Hausdorff distance measures the maximum distance between the AI boundary and the human boundary, right?

Maya: Correct. And an average error of 0.54 mm is tiny, especially when you are looking at large vessels like the aorta. For the centerline accuracy, the average Hausdorff distance was 0.59mm. They also measured the reproducibility of generating ground truths, which showed an average Dice coefficient of 0.95.

Sam: What about the actual measurements the software spits out? If a surgeon is deciding whether to operate based on the diameter of an aneurysm, they need that measurement to be rock solid.

Maya: The document states that the AM measurements were compared to measurements taken from approved ground truth segmentations using a validated technique. The mean absolute error was found to be 0.22 cm.

Sam: 0.22 cm. That is pretty tight. The summary notes this compares favorably with the reference device, which is listed earlier as Rapid RV/LV, cleared under K223396.

Maya: Exactly. They used Rapid RV/LV specifically as a reference for measurement accuracy.

Sam: Okay, the numbers look great, 100% agreement, sub-millimeter boundary differences. But AI in the real world is messy. What are the limitations? When does this software fail or refuse to work?

Maya: The Precautions and Exclusions section is quite detailed. First, it warns that series containing excessive patient motion or metal implants may impact module output quality.

Sam: That is a major limitation for older adults who might have spinal hardware or hip replacements near the iliac arteries. What else will the module outright refuse to process?

Maya: The software will not process series acquired with cone-beam CT scanners, also known as c-arm CT. It excludes series that are non-axial or axial oblique greater than 5 degrees. It also rejects series containing improperly ordered or missing slices where the gap is larger than 3 times the median inter-slice distance, which they note can happen as a result of manual correction by an imaging technician.

Sam: So if a tech deletes some blurry slices manually and creates a gap, the AI might just reject the whole scan.

Maya: Exactly. It also requires at least 3cm of target anatomical zones to work. Beyond that, there are strict dataset parameters. It excludes scans with an in-plane X and Y field of view less than 160mm, and a Z field of view, meaning cranio-caudal coverage, less than 144 mm. The in-plane pixel spacing must be between 0.3 mm and 1.0 mm.

Sam: And you mentioned slice thickness earlier. What were the exact limits?

Maya: It excludes an inter-slice distance of less than 0.3 mm or greater than 3 mm, and a slice thickness greater than 3 mm. Finally, it excludes data acquired at an x-ray tube voltage less than 70kVp or greater than 150kVp. This applies to single energy, dual energy, or virtual monochromatic datasets.

Sam: So hospital IT and radiology directors need to verify that their standard aortic CT protocols actually fit tightly inside these parameters. If your standard slice thickness is 5 mm, this software will not even look at the scan.

Maya: Precisely. And for IT departments, the summary provides some good details on implementation and security. The software can be installed on-premises within a customer infrastructure behind their firewall, or in a hybrid on-premises/cloud configuration. It accepts DICOM images and returns the processed images back to the source modality or PACS.

Sam: What about cybersecurity? Hospitals are getting hammered by ransomware right now.

Maya: The document states that the software conforms to several standards, including ISO 14971:2019 for risk management and IEC 62304:2015 for the software lifecycle. Specifically for cybersecurity, it says it has been designed to meet requirements using design Vulnerability Assessments, PEN Testing, and importantly, SBOMs.

Sam: SBOMs, Software Bill of Materials. That is huge for hospital procurement teams today. They want to know exactly what code is inside the black box they are buying. Are there any other standards listed in the filing?

Maya: Yes, they also list IEC 62366-1:2015 for usability engineering, and NEMA PS 3.1 - 3.20, which is the standard for Digital Imaging and Communications in Medicine, or DICOM. It shows the device is built against recognized technical frameworks.

Sam: That is reassuring. But returning to the clinical side, there is always a catch with AI. What is the final disclaimer from the FDA?

Maya: The most critical clinical warning in the document is standard but vital: Rapid Aortic Measurements is an aid to physician decision making. Its results are not intended to be used on a stand-alone basis for clinical decision-making or otherwise preclude clinical assessment.

Sam: The classic FDA safety net. The AI can draw the lines and give you the measurements, but the doctor still owns the final diagnosis. It is fascinating to see a company take the infrastructure they built for rapid stroke triage and apply it directly to aortic emergencies.

Maya: It is a logical expansion. The basic tasks, removing bone, segmenting vessels, computing centerlines, are technologically similar, even if the anatomy and the technical tolerances are different.

Sam: Alright, let us recap. iSchemaView has received 510(k) clearance for Rapid Aortic Measurements, moving their AI capabilities into the chest and abdomen. The performance testing showed incredibly tight accuracy, 100% agreement from clinical experts, and a mean absolute error of 0.22 cm, despite some mathematical typos in the dataset description.

Maya: But buyers must ensure their scanner protocols match the strict input requirements, keeping slice thickness to 3 mm or less and avoiding excessive motion artifacts.

Sam: Exactly. And for our listeners, please remember that this podcast is for informational purposes only and is absolutely not medical advice. You can find a link to the full FDA 510(k) summary in the show notes to read the source document for yourself.

Maya: Thank you for tuning in to this Company Spotlight. We will catch you next time on AI in Medicine - Smart Summaries.