Company Spotlight: Bunkerhill Contrast CAC - Opportunistic Screening on Routine Scans

Maya: Welcome to AI in Medicine - Smart Summaries. I'm Maya, and just a quick reminder right at the top: our voices are AI-generated. Today, we are looking at a clearance that caught my eye because it hit a sensitivity of 0.957 for finding coronary artery calcium on routine scans that were never actually meant to look for it.

Sam: Exactly. I'm Sam. Today we are doing a Company Spotlight episode on a public document: the FDA 510(k) summary for a product called Bunkerhill Contrast CAC, published by the U.S. Food and Drug Administration on March 6, 2026. Just to be totally clear, this is a summary of the FDA's public clearance document, not a reading of company marketing material.

Maya: The single most interesting finding here is how the software bridges a gap in radiology. It automatically analyzes non-gated, contrast-enhanced chest CT images to detect and estimate the quantity of coronary artery calcification. That contrast-enhanced part is the key.

Sam: Right, because normally, looking for coronary artery calcium requires a very specific type of scan, right?

Maya: Yes, standard calcium scoring is usually done on non-contrast scans. In fact, if we look at the primary predicate device listed in this document, which is called the iCAC Device cleared under K230223, that one automatically analyzes non-gated, non-contrast chest CT images. Bunkerhill Contrast CAC moves this capability to contrast-enhanced scans.

Sam: Which is huge from a practical standpoint. If a patient comes in and gets a contrast CT for an entirely different reason, say to look at their lungs or rule out a pulmonary embolism, this software can opportunistically check their heart for calcium at the same time. Who exactly is this cleared for?

Maya: The indications for use say it is for adult patients aged 30 years and above. And the software automatically analyzes these chest CT images collected during clinical care, and outputs a region of interest, which is intended for informational purposes only, as well as a quantification of the detected calcium.

Sam: Wait, really? Age 30 and above? That seems quite young for standard cardiovascular screening. But again, since it's opportunistic, it makes sense that they want to capture any early signs of calcification if the scan is already being done.

Maya: Exactly. The document notes that the patient population for the predicate device was also patients aged 30 years and above, so they are maintaining that exact same age threshold. The reference device they used, the Bunkerhill AVC, was for patients aged 40 years and above, but for the coronary artery calcium, they stuck with 30.

Sam: Let's dig into how it actually works. The summary says this is a software as a medical device, or SaMD, product. What exactly is under the hood?

Maya: The summary states that both the predicate and the subject device use deep-learning algorithms to identify the presence of calcification and estimate calcification. They both analyze non-gated chest CT images sent to the software in DICOM format.

Sam: DICOM being the standard medical imaging format. And how does the doctor actually see this information? Does it pop up on their screen and interrupt what they are doing?

Maya: No, workflow integration was clearly a focus here. The document says the output is made available to the physician on-demand as part of his or her standard workflow. The device-generated quantification can be viewed in the patient report at the discretion of the physician.

Sam: So it's there if they want it, but it doesn't replace their standard view.

Maya: Exactly. The physician also has the option of viewing the device-generated calcium region of interest in a diagnostic image viewer. The summary specifically states the output in no way replaces the original patient report or the original non-gated, contrast-enhanced CT scan. Both are still available to be viewed and used at the discretion of the physician.

Sam: That's a very standard, safe FDA lane. It's a support tool. In fact, the document says results are not intended to be used on a stand-alone basis and are solely intended to aid and provide information to the physician. They mention further action taken on a patient should only come at the recommendation of the physician.

Maya: Right. But there are some fascinating technological differences between this new device and the predicate device that health-system buyers and AI builders should pay attention to. The biggest one is how they quantify the calcium.

Sam: Oh? How do they do it?

Maya: The predicate device outputs Agatston Units using a deterministic formula based on raw segmentation maps. But the subject device, Bunkerhill Contrast CAC, outputs volume in mm3 using a deterministic mathematical formula based on the raw segmentation maps. It also does not have a categorical output.

Sam: Wait, it doesn't use Agatston scores? That's the gold standard metric every cardiologist uses for calcium scoring. Why the shift to purely volume in mm3?

Maya: The document does not explicitly explain the medical reason for dropping Agatston units, it only notes that it is a technological difference. However, it does mention that the subject device does not have a categorical output as coronary artery calcification, and that this aspect is most similar to their reference device, Bunkerhill AVC.

Sam: That makes sense. If you are building AI in this space, you have to realize that switching from non-contrast to contrast-enhanced scans changes the imaging physics. You might have to change your output metric to maintain accuracy, which is exactly what they did by moving to mm3 volume. What about the visual output? How does the doctor see the calcium?

Maya: The predicate device displays estimated calcium segmentation. But this subject device shows a visual output in the form of a Region of Interest, or ROI. The summary says the rationale behind this was to ensure that the interpreting physician is able to visualize the underlying calcification.

Sam: So instead of painting over the calcium exactly, it draws a region of interest around it. What if the AI is wrong? What if it flags something that isn't calcium?

Maya: They built in a safeguard for that. The document says the user has the option to delete the visual outputs if they do not agree with the results in both the subject and predicate device.

Sam: That is a great feature for clinical adoption. Doctors hate it when an AI burns a permanent, incorrect marker onto an image. Giving them the power to delete it keeps the physician in control.

Maya: Exactly. Now, let's talk about the performance testing, because that is where we get the real numbers on how well this actually works. The safety and performance were validated in a stand-alone retrospective study for detection, localization, and agreement of the device output compared to the established ground truth.

Sam: Retrospective, meaning they looked at past scans, not a prospective trial where they follow patients in real-time. Where did they get the data?

Maya: The pivotal dataset was curated from multiple sites across three geographical regions in the United States. And they hit their co-primary endpoints. The observed Sensitivity was 0.957, with a confidence interval of 0.924 to 0.991.

Sam: So it's highly sensitive. It finds the calcium when it's there. What about false positives? What was the specificity?

Maya: The observed Specificity was 0.929, with an interval from 0.882 to 0.976. They also reported an observed Precision of 0.814, and an observed Recall of 0.755.

Sam: Those are solid numbers for a tool meant to operate opportunistically in the background. But when they estimate the actual volume of the calcium, how close are they to the ground truth? You mentioned mm3 earlier.

Maya: They used a Bland Altman agreement to measure that. The observed bias in the pivotal study was 25.41 mm3. The lower and upper limits of agreement were -249.13 mm3 and 198.32 mm3, respectively.

Sam: A bias of 25.41 mm3 is pretty small, meaning on average, the AI's volume estimate is very close to the ground truth. The document itself says, 'The low mean difference demonstrates that Bunkerhill Contrast CAC has a bias of low magnitude.' Though those limits of agreement, going from -249.13 to 198.32, show there can be some individual variation.

Maya: Yes, but they state that Bunkerhill Contrast CAC met the mean difference and limits of agreement acceptance criteria, thus satisfying all the endpoints of the pivotal testing study. And they didn't stop there. They also did an additional bridging study.

Sam: What does a bridging study do in this context?

Maya: The bridging study compared the volume generated by the Contrast CAC device on a contrast-enhanced image to the calcification volume on the same patient calculated on a non-contrast image. This was to further demonstrate the validity of detecting and quantifying calcium in line with its intended use.

Sam: That is a very smart study design. If you are claiming you can do this on contrast scans, you need to prove it gives you similar results to the old way of doing it without contrast on the exact same person. How big was this study?

Maya: They included 50 cases in the bridging study. The study met the predefined acceptance criteria with a mean bias of -4.36 mm3 and limits of agreement of -76.63 mm3 to 67.91 mm3.

Sam: 50 cases. That is a somewhat limited data set for the bridging study, but the mean bias of -4.36 mm3 is incredibly tight. It shows that whether the scan has contrast or not, the volume calculation remains very stable for those 50 patients.

Maya: Exactly. Based on all of this, the FDA determined the device is substantially equivalent to the predicate devices. The document concludes that the minor differences in indications do not alter the intended diagnostic use of the device and do not affect its safety and effectiveness when used as labeled.

Sam: It's worth reminding our listeners about the limitations here. The clearance is based on a stand-alone retrospective study. There is no prospective clinical trial data presented in this summary showing how this impacts patient outcomes in the real world. And as a Class II device cleared under the 510(k) pathway, the bar is substantial equivalence to an existing device, not necessarily proving it saves lives.

Maya: That is exactly right. The document specifically notes that CDRH does not evaluate information related to contract liability warranties, and that FDA's issuance of a substantial equivalence determination does not mean the FDA has made a determination that the device complies with other requirements of the Act.

Sam: For companies building AI, the big takeaway from this document is how BunkerHill Health navigated the transition from non-contrast to contrast imaging. They changed their output from Agatston categories to a continuous volume metric in mm3, and they ran a specific 50-case bridging study to prove to the FDA that the contrast didn't break their volume calculations.

Maya: And for clinicians and health-system leaders, the takeaway is the expansion of opportunistic screening. If your hospital runs this software, an adult over 30 getting a contrast chest CT for a lung issue or to rule out a pulmonary embolism could automatically have their coronary calcium quantified and flagged for you to review on-demand, without ordering a separate cardiac scan.

Sam: As long as the doctor remembers that they are the final check, and they have the power to delete that region of interest if the AI gets it wrong.

Maya: Exactly. Let's do a crisp recap. We covered the FDA 510(k) summary for Bunkerhill Contrast CAC. It is an AI software cleared to opportunistically detect and quantify coronary artery calcium volume on non-gated, contrast-enhanced chest CTs for adults 30 and older. In testing, it achieved a sensitivity of 0.957 and a specificity of 0.929.

Sam: Remember, the original FDA summary document is linked in the show notes for you to read yourself. And as always, this podcast is for informational purposes only and is not medical advice. Thanks for listening.