Purpose of Review Imaging derived fractional flow reserve (FFR) is rapidly evolving beyond conventional computational fluid dynamics (CFD) based pipelines toward machine learning (ML), deep learning (DL), and physics informed approaches that enable fast, wire free, and scalable functional assessment of coronary stenosis. This review synthesizes recent advances in CT and angiography based FFR, with particular emphasis on emerging physics informed neural networks and neural operators (PINNs and PINOs) and key considerations for their clinical translation. Recent Findings ML/DL approaches have markedly improved automation and computational speed, enabling prediction of pressure and FFR from anatomical descriptors or angiographic contrast dynamics. However, their real-world performance and generalizability can remain variable and sensitive to domain shift, due to multi-center heterogeneity, interpretability challenges, and differences in acquisition protocols and image quality. Physics informed learning introduces conservation structure and boundary condition consistency into model training, improving generalizability and reducing dependence on dense supervision while maintaining rapid inference. Recent evaluation trends increasingly highlight deployment oriented metrics, including calibration, uncertainty quantification, and quality control gatekeeping, as essential for safe clinical use. Summary The field is converging toward imaging derived FFR methods that are faster, more automated, and more reliable. While ML/DL offers substantial efficiency gains, physics informed frameworks such as PINNs and PINOs may provide a more robust balance between speed and physical consistency. Prospective multi center validation and standardized evaluation will be critical to support broad and safe clinical adoption.
Coronary artery disease (CAD) remains a leading cause of morbidity and mortality worldwide. A central clinical challenge is to distinguish anatomically visible stenosis from functionally significant lesions that truly cause ischemia, particularly for intermediate disease where visual assessment is subjective and variable. Pressure-wire fractional flow reserve (FFR), measured under maximal hyperemia, is widely regarded as the invasive reference standard for lesion-specific physiological assessment and for guiding revascularization decisions [1][2].
Despite its clinical value, routine pressure-wire FFR is not universally adopted in daily practice. Practical barriers include wire manipulation, the need for hyperemic agents, additional procedure time and cost, and patient-or operator-related constraints [3]. These limitations have motivated the development of wire-free, imaging-derived physiological assessment-aiming to provide actionable functional information with reduced procedural burden while preserving lesion-level decision support.
In this review, we focus on non-invasive imaging-derived approaches that infer coronary hemodynamics and functional significance from routinely acquired imaging, with an emphasis on methods that can realistically integrate into clinical workflows and provide reliable outputs under heterogeneous image quality, acquisition protocols, and lesion morphologies.
Imaging-derived functional assessment leverages anatomical and/or contrast-flow information to estimate pressure drop and physiological significance along the coronary tree. Among available modalities, clinical translation and large-scale evidence have been most prominent for computed tomography coronary angiography (CCTA) and invasive coronary angiography (ICA), each with distinct strengths and constraints.
CCTA provides a patient-specific 3D representation of the coronary anatomy and plaque burden, enabling physiologic assessment across the vascular tree. CT-derived FFR has demonstrated diagnostic value relative to invasive FFR and has been positioned as a noninvasive strategy to improve patient triage beyond anatomy alone [3][4][5][6]. However, real-world feasibility can be limited by image quality and reconstruction challenges (e.g., motion artifacts, noise, heavy calcification or stent blooming), as well as sensitivity to segmentation and centerline uncertainties. Consequently, practical adoption depends not only on diagnostic accuracy but also on computability, quality control, and robust handling of imperfect inputs [7][8]. Cardiac CT artifact mechanisms and mitigation strategies (including blooming and other physics-driven artifacts) have been comprehensively discussed elsewhere [9][10].
Angiography-based functional assessment (ICA-derived physiology): ICA is routinely performed for diagnosis and procedural planning in the catheterization laboratory. A major clinical motivation for ICA-derived physiology is to provide wire-free functional information within the cath-lab time budget, supporting immediate decision-making without advancing a pressure wire. Quantitative flow ratio (QFR) and related approaches have reported strong agreement with invasive FFR in validation studies, and randomized evidence has supported QFR-guided strategies [11][12]. In practice, angiography-derived methods are highly dependent on acquisition conditions, including adequate projection separation, minimal foreshortening/overlap, and stable contrast opacification; real-world studies and head-to-head analyses have highlighted non-computability drivers and practical requirements. These workflow constraints make explicit quality control and failure-mode management integral to real-world performance [13][14].
To synthesize a rapidly growing literature, we organize imaging-derived FFR methods into three methodological families that reflect different trade-offs among fidelity, speed, interpretability, and robustness:
(i) Computational Fluid Dynamics (CFD) and physics-based modeling: Physics-based approaches estimate pressure and flow by solving governing equations on image-derived coronary geometries. They offer mechanistic consistency and interpretability, but clinical feasibility is often constrained by computational cost and sensitivity to physiological assumptions, most notably boundary conditions and microvascular modeling [15][16][17][18].
(ii) Machine learning and deep learning: ML-based approaches replace part or all of the physics pipeline with data-driven inference, enabling near-instant computation and high automation. Their core challenge is reliability under dataset shift: heterogeneous scanners, protocols, image quality, and lesion morphology can degrade generalization. This has motivated increasing attention to calibration, uncertainty estimation, and deployment-aware risk management [19][20][21][22].
(iii) Physics-informed learning (PINNs and neural operators): Physics-informed learning aims to bridge CFD and ML by embeddin
This content is AI-processed based on open access ArXiv data.