HeartFlow Analysis

HeartFlow Analysis is an AI-based software device that analyzes previously acquired computed tomography (CT) DICOM images to assess coronary artery disease in adult patients ages 22 and older. It provides anatomical data, plaque localization and characterization, and calculates fractional flow reserve (FFR_CT), a physiological simulation derived from simulated pressure, velocity, and blood flow information. The device is intended to support qualified clinicians in evaluating and assessing the risk of coronary artery disease in conjunction with clinical history and other diagnostic tests.

HeartFlow Analysis is a software medical device for quantitative and qualitative analysis of coronary CT angiography (cCTA) using deep learning algorithms trained since 2015. The core technology remains unchanged from the primary predicate (K213857); key technological characteristics are identical between subject and predicate devices, including DICOM compatibility, internet-connected delivery of graphic and text results, and non-invasive operation. The subject device (version 4.0) introduces a retrained and validated plaque analysis algorithm with new plaque visualization and quantification capabilities, plus automation of internal quality controls, compared to the predicate (version 3.18).

Not stated in this summary. The document references validation studies including stress testing and repeatability testing, but does not cite specific consensus standards such as ISO, IEC, or ASTM specifications.

HeartFlow Analysis is substantially equivalent because it maintains the same intended use (with clarified patient population age 22+ and terminology change from 'identification' to 'localization' of plaque), identical operating principles, and similar technological characteristics as the legally marketed predicate K213857. The plaque algorithm modifications are supported by extensive validation on 60,555 ground-truth areas from 583 lesions across 100 patients and demonstrate statistically significant improvements in sensitivity (0.151 superiority, p<0.0001) and quantification accuracy while maintaining equivalent DICE similarity (0.8 vs. 0.8). Risk assessment confirmed the modifications do not raise new safety or effectiveness questions, and the Predetermined Change Control Plan ensures future algorithmic improvements maintain established performance thresholds.