Optimizing Non-invasive Fractional Flow Reserve Estimation with Machine Learning-Enhanced 1D Hemodynamic Modeling.

Area of Science:

• Cardiovascular Computational Modeling
• Medical Machine Learning
• Diagnostic Imaging Analysis

Background:

• Patient-specific computational models correlate well with invasive fractional flow reserve (FFR) measurements for diagnosing coronary ischemia.
• Current FFR modeling is computationally intensive and underutilizes available patient data, hindering clinical use.
• Existing methods often rely on complex pulsatile flow assumptions.

Purpose of the Study:

• To develop a computationally efficient, hybrid coronary angiography-based approach for patient-specific FFR estimation.
• To integrate physics-based modeling with machine learning (ML) for improved FFR prediction accuracy.
• To leverage routinely available clinical data more effectively in FFR modeling.

Main Methods:

• A hybrid framework combining steady-state flow assumptions with ML feedback loop was developed.
• Physics-based modeling was integrated with an ML component to refine FFR predictions.
• A retrospective, two-center cohort of 132 patients with 132 coronary lesions was used for evaluation.

Main Results:

• Steady-state models accurately captured essential hemodynamic patterns, closely matching pulsatile model predictions.
• The ML refinement significantly improved diagnostic accuracy.
• Achieved sensitivity: 83.3%, specificity: 100.0%, PPV: 100.0%, NPV: 88.2%, overall precision: 92.6%.

Conclusions:

• The hybrid approach offers a robust and clinically viable method for accurate patient-specific FFR estimation.
• Simplified steady-state flow assumptions are effective for hemodynamic pattern capture in FFR modeling.
• Combining efficient computational modeling with ML-driven refinement enhances diagnostic performance for coronary ischemia.