Invasive fractional-flow-reserve prediction by coronary CT angiography using artificial intelligence vs....

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary computed tomography angiography (CCTA) with non-invasive fractional flow reserve (FFR) assesses lesion-specific ischemia.
• Current guidelines suggest CCTA-FFR for stable chest pain patients with intermediate-grade stenoses.
• Invasive FFR remains the reference standard for assessing ischemia.

Purpose of the Study:

• To compare a novel CCTA-based artificial intelligence deep-learning model for FFR prediction (FFR_AI) against computational fluid dynamics CT-derived FFR (FFR_CT).
• To evaluate the diagnostic performance of FFR_AI and FFR_CT in identifying intermediate-grade coronary stenoses using invasive FFR as the reference standard.

Main Methods:

• A retrospective proof-of-concept study included 37 patients with 39 intermediate-grade stenoses.
• The FFR_AI model was trained using CCTA images from 500 vessels.
• FFR_AI and FFR_CT were compared for sensitivity, specificity, PPV, NPV, and diagnostic accuracy in predicting FFR ≤ 0.80.

Main Results:

• FFR_AI demonstrated a diagnostic accuracy of 85% (33/39) for predicting FFR ≤ 0.80.
• FFR_CT showed a diagnostic accuracy of 77% (30/39) for predicting FFR ≤ 0.80.
• No significant difference in diagnostic accuracy was observed between FFR_AI and FFR_CT (p=0.12).

Conclusions:

• FFR_AI performed similarly to FFR_CT in predicting significant coronary stenoses (FFR ≤ 0.80).
• FFR_AI shows potential as a non-invasive imaging tool for guiding therapeutic decisions in patients with intermediate-grade stenoses.