DEEPVESSEL FFR

DEEPVESSEL FFR is a software device that utilizes deep learning neural networks to analyze static coronary CTA images for non-invasive estimation of Fractional Flow Reserve (DVFFR) in patients with suspected coronary artery disease. Its methodology involves automated 3D coronary tree segmentation followed by AI-driven physiological simulation, providing quantitative and qualitative data to aid clinicians in the functional assessment of coronary lesions.

DEEPVESSEL FFR is a coronary physiological simulation software designed for the quantitative and qualitative analysis of coronary artery disease in clinically stable symptomatic patients using previously acquired Computed Tomography (CT) DICOM data. The software employs deep learning neural networks to compute DVFFR, a CT-derived Fractional Flow Reserve measurement, from static coronary CTA images. Its methodology involves a semi-automatic process, first generating a 3D model of the coronary artery tree using deep learning-based segmentation algorithms, which can be manually refined by qualified image analysts. Subsequently, a deep learning framework, comprising a multi-layer perceptron network (MLP) and a bidirectional multi-layer recursive neural network (BRNN), utilizes these segmented models and CTA images to estimate semi-continuous FFR values along the coronary artery centerlines, aiming to support the functional evaluation of CAD. The clinical effectiveness of DEEPVESSEL FFR was validated in a multi-national, multicenter study involving 244 patients and 311 target vessels, comparing its DVFFR measurements against invasive FFR. On a per-vessel basis, the software demonstrated a sensitivity of 86.9% (95% CI: 80.6%–92.7%) and a specificity of 86.7% (95% CI: 82.0%–91.1%), with a diagnostic accuracy of 86.8%. Patient-level analysis showed similar strong performance, with sensitivity at 87.4% and specificity at 83.7%. The output of the software includes a PDF report with detailed DVFFR assessments, branch-by-branch visualizations, and a 3D DVFFR tree model. These results are intended to assist qualified clinicians in evaluating coronary arteries, in conjunction with other clinical data and professional judgment.

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 deep learning algorithm in DEEPVESSEL FFR vary across different CTA scanner manufacturers and image acquisition protocols not explicitly included in the validation study?; What is the computational time required for DEEPVESSEL FFR to process a typical coronary CTA dataset and generate results, and how does this compare to other non-invasive FFR estimation methods?; The document mentions "manual edits to the 3D coronary artery tree segmentations." How frequently are such edits required by qualified image analysts, and what is the inter-observer variability in these manual corrections and their impact on final DVFFR values?; Given the use of deep learning, how robust is DEEPVESSEL FFR to variations in patient anatomy (e.g., presence of stents, bypass grafts, or severe calcification) that might not have been extensively represented in the training dataset?; What are the specific imaging, structural, and functional characteristics encoded by the deep learning neural networks, and how does the system weigh these characteristics in estimating DVFFR?