CAC (gated) Algorithm

The CAC (gated) Algorithm by BunkerHill Health is a software medical device designed to quantify and categorize coronary artery calcium from ECG-gated cardiac CT scans using deep learning. Its purpose is to provide physicians with adjunctive, on-demand information, including Agatston scores and calcium segmentation, to assist in assessing cardiovascular risk in patients aged 30 and above without replacing standard diagnostic procedures.

The CAC (gated) Algorithm is a software medical device developed by BunkerHill Health, designed to assist physicians in estimating the presence and quantity of coronary artery calcium (CAC) in patients aged 30 years and older. It functions by automatically analyzing non-contrast, electrocardiogram (ECG) gated cardiac computed tomography (CT) images. The primary aim is to provide supplementary information to clinicians during their review of a patient's case, integrated into their standard workflow, to aid in the assessment of coronary artery calcification. The device utilizes deep learning algorithms to interface with compatible CT systems, processing DICOM-formatted, non-contrast ECG-gated cardiac CT images. It localizes, quantifies, and categorizes coronary artery calcification. Key outputs include the categorization of the overall coronary artery calcium burden into four Agatston score ranges (0, 1-100, 101-400, >400 Agatston units), an exact overall Agatston score, and exact Agatston scores for individual coronary arteries (Left Main, Left Anterior Descending, Left Circumflex, and Right Coronary Artery). Additionally, it generates a DICOM segmentation map of the estimated calcium burden for informational viewing; however, these outputs are intended to supplement, not replace, the original CT scans or the physician's comprehensive report. A retrospective validation study demonstrated strong performance, with a Cohen’s weighted kappa of 0.92 for agreement with ground truth in classifying cardiovascular disease risk based on Agatston score categories.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the deep learning algorithm's performance vary with different CT scanner models, slice thicknesses (up to 3mm), and acquisition protocols used for the input ECG-gated images?; What is the inter-observer variability when physicians interpret the device-generated Agatston scores and segmentation maps compared to manual or other semi-automated methods?; Beyond the four Agatston score categories, how does the device's exact Agatston score output for individual vessels (LM, LAD, LCx, RCA) correlate with clinical outcomes or aid in more granular risk stratification?; What were the specific characteristics of the 'ground truth' used in the retrospective validation study, and how was it established (e.g., expert consensus, specific software)?; How does the device handle artifacts or challenging image quality scenarios in ECG-gated CT scans, and what is its reported failure rate or need for manual correction in such cases?