On-Site Computed Tomography-Derived Fractional Flow Reserve Using a Machine-Learning Algorithm　- Clinical...

Area of Science:

• Cardiovascular Imaging
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Coronary artery disease (CAD) diagnosis relies on invasive fractional flow reserve (FFR) measurements.
• Coronary computed tomography angiography (CTA) can identify stenosis but often requires functional testing for confirmation.

Purpose of the Study:

• To evaluate the diagnostic capability of on-site CT-FFR for detecting hemodynamically significant CAD.
• To compare the diagnostic performance of CT-FFR with standard coronary CTA interpretations.

Main Methods:

• Retrospective review of 74 patients with coronary CTA and invasive FFR measurements.
• CT-FFR computed using a machine learning (ML) algorithm on 91 coronary vessels.
• Comparison of CT-FFR results with invasive FFR and CTA-based stenosis grading.

Main Results:

• Good correlation between CT-FFR and invasive FFR (r=0.786, P<0.001).
• CT-FFR demonstrated superior diagnostic performance (AUC=0.907) compared to CTA stenosis assessment (AUC=0.595-0.603).
• On-site CT-FFR analysis improved per-patient diagnostic accuracy from 66% to 85% by correcting or confirming standard CTA classifications.

Conclusions:

• On-site CT-FFR using an ML algorithm offers strong diagnostic performance for significant CAD.
• Coronary CTA combined with CT-FFR provides high diagnostic value for selected patients in clinical practice.