HealthCCS

HealthCCS is a software tool designed to automatically quantify coronary artery calcification from cardiac CT scans using a voxel density threshold and a CNN-based probability assessment to calculate an Agatston-equivalent risk score. The system aims to provide a non-invasive method for physicians to assess cardiovascular risk, with performance validated by agreement with radiologist-derived ground truth and demonstrating good reproducibility.

HealthCCS is a non-invasive, automatic post-processing software tool designed to evaluate calcified plaques in the coronary arteries from pre-existing non-contrast, ECG-gated/triggered cardiac or chest CT scans for patients over 20 years old. Its primary purpose is to aid physicians in assessing the risk of coronary artery disease by identifying and quantifying coronary artery calcification. The software operates by first identifying calcification based on a voxel density exceeding a predefined threshold of 130 Hounsfield Units (HU). Subsequently, a Convolutional Neural Network (CNN)-based probability threshold is applied to ascertain the likelihood that the identified calcium is located within the coronary arterial distribution. The HealthCCS system then calculates a Coronary Artery Calcium score using the Agatston method, averaging scores from multiple series if present, to provide a single Agatston-equivalent score for the study. The HealthCCS device generates patient reports that include a 4-category Agatston-equivalent risk score, which physicians can use for further analysis and to guide treatment decisions, particularly noting that patient management for low scores (0-10) relies on physician judgment and may necessitate additional testing. Performance validation was conducted through a retrospective study involving 249 scans, comparing HealthCCS's risk categorization against a ground truth established by three radiologists using a reference PACS device. This study demonstrated an adequate overall agreement of 0.89 (95% CI: 0.85-0.92) and good reproducibility, with identical Agatston equivalent scores across three readings for 150 studies. The device was found to provide accurate calcium scoring, meeting requirements for accuracy, precision, and reproducibility.

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 performance of the CNN-based probability threshold in HealthCCS for localizing coronary calcium compare to manual or semi-automatic methods in terms of accuracy and inter-observer variability, especially in complex anatomies or low-quality scans?; What is the clinical impact of HealthCCS providing only a total Agatston-equivalent score versus predicate devices that offer per-artery calcium scores, particularly in guiding localized interventions or more detailed risk stratification?; The document mentions an average Agatston equivalent score if multiple series are present. What is the rationale and validation for averaging scores from different series, and how does this approach affect the final risk categorization compared to selecting a single representative series?; Given the reliance on physician judgment for scores 0-10, what specific guidance or data does HealthCCS provide to support clinicians in these borderline cases, and how does its performance compare to other tools in this low-score range?; How does the spatial threshold of 1.5 mm² used by HealthCCS, compared to 1 mm² in the predicate, affect the sensitivity and specificity of calcification detection, particularly for smaller plaques, and what was the justification for this difference?