CAC Software

The Imbio CAC Software is a machine learning-driven, non-invasive post-processing solution that analyzes non-contrast thoracic CT images to automatically segment and quantify coronary artery calcifications. It provides clinicians with objective metrics including Agatston score, calcified lesion mass, and volume to aid in the assessment of cardiovascular disease risk in adult patients, demonstrating high accuracy in risk categorization against expert-annotated data.

The Imbio CAC Software is a non-invasive, medical image post-processing tool designed to assist in evaluating coronary artery calcification, a significant indicator of coronary artery disease risk. Employing machine learning algorithms, the software automatically analyzes non-contrast thoracic computed tomography (CT) images from adult patients aged 29 years and older. Its primary function is to segment and quantify calcified plaques within the coronary arteries. The system outputs a comprehensive summary report that includes the Agatston score, arterial age, and specific metrics for calcified lesion mass and volume, detailed for the entire heart and for individual coronary arteries (RCA, LAD, LCx). Additionally, it generates annotated DICOM images visualizing the calcification segmentations for informational review, intended to be integrated into the standard DICOM viewing workflow to support physicians, though it is not designed for standalone diagnostic decisions or to supplant clinical judgment in CT image assessment. The Imbio CAC Software operates as a command-line executable that processes previously acquired thoracic CT DICOM datasets. It is compatible with images derived from a range of CT scanners and supports both low and standard dose scans, as well as cardiac-gated and non-cardiac-gated acquisitions with slice thicknesses up to 3mm, using a default Hounsfield Unit threshold of 130 for calcium detection. The software's efficacy was evaluated in a retrospective, multi-center study involving 500 anonymized chest CT series. The key performance outcome demonstrated a high level of agreement with ground truth data (derived from expert annotations using an established semi-automated software) in categorizing cardiovascular disease risk. Specifically, it achieved a Cohen’s kappa of 0.907 (95% CI 0.895, 0.920) for a 5-category risk assessment, successfully meeting its predefined acceptance criteria.

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 machine learning algorithm differentiate between true coronary artery calcifications and other high-density structures or artifacts, such as surgical clips or aortic valve calcification, in non-contrast thoracic CT scans?; What was the variability in the software's performance (e.g., Agatston score accuracy, segmentation quality) across the different CT scanner manufacturers (GE, Siemens, Imatron, Philips) and varying acquisition parameters (e.g., slice thickness up to 3mm, radiation dose) included in the 500-case validation study?; The validation study included an equal split of ECG-gated and non-gated acquisitions; were there statistically significant differences in the accuracy of calcification quantification or risk categorization between these two acquisition types, and how does the software account for potential motion artifacts in non-gated scans?; Beyond the Agatston score and 5-category risk assessment, how do the reported calcified lesion mass and volume metrics correlate with long-term patient outcomes, progression of coronary artery disease, or response to therapy compared to traditional risk factors or Agatston score alone?; What specific image pre-processing steps are incorporated into the Imbio CAC Software prior to the machine learning analysis to normalize image quality and ensure robustness across diverse datasets from various clinical sites and CT scanners?