Imaging-Derived Coronary Fractional Flow Reserve: Advances in Physics-Based, Machine Learning, and Physics-Informed...

Area of Science:

• Cardiovascular Imaging
• Medical Artificial Intelligence
• Computational Physiology

Background:

• Fractional flow reserve (FFR) is crucial for assessing coronary artery stenosis severity.
• Conventional FFR measurement requires invasive pressure wire assessment.
• Non-invasive imaging-derived FFR methods are emerging to overcome limitations.

Purpose of the Study:

• To review recent advancements in imaging-derived FFR, focusing on machine learning (ML), deep learning (DL), and physics-informed approaches.
• To discuss the clinical translation challenges and considerations for these novel FFR techniques.
• To highlight the convergence towards faster, automated, and reliable FFR assessment.

Main Methods:

• Review of recent literature on CT and angiography-based FFR.
• Emphasis on physics-informed neural networks (PINNs) and neural operators (PINOs).
• Discussion of ML/DL applications for predicting pressure and FFR from imaging data.

Main Results:

• ML/DL methods enhance automation and speed for FFR prediction.
• Physics-informed learning improves generalizability and reduces reliance on extensive supervision.
• Real-world performance of ML/DL can be variable due to data heterogeneity and acquisition differences.

Conclusions:

• The field is moving towards more efficient and reliable imaging-derived FFR.
• Physics-informed frameworks offer a balance between speed and physical consistency.
• Multi-center validation and standardized evaluation are essential for clinical adoption.