Background. To diagnose, assess severity and guide treatment, currently, most patients with ischaemic heart disease undergo 2-dimensional, x-ray angiography (CA). During CA, optical coherence tomography (OCT) can provide high-resolution, intravascular images of the lumen. Coronary physiology is used to assess various pressure- and flow-based metrics. CA, OCT and physiology have a class 1A indication in international guidelines yet are rarely used together.
Aims. To develop a novel, prototype computational method that fused all three assessments generating high-resolution, 3D, bifurcation anatomy models, to simulate physiology.
Methods. The novel toolchain fused 3D arterial centrelines (paired CA-images) and lumen contours (OCT), creating three reconstruction sets: CA-only and fused CA-OCT single vessels and CA-OCT branched vessels. Novel approaches to centreline torsion correction, bifurcation merging, rotational OCT optimisation and fusion of vessel surfaces were developed. CFD simulations were performed computing virtual fractional flow reserve (vFFR), absolute flow and microvascular resistance. Anatomical and physiological results were compared with state-of-the-art modelling techniques and against in vivo data.
Results. The prototype model successfully reconstructed twenty coronary arteries and eight bifurcations, from twelve different patients with stable coronary artery disease, from two tertiary cardiology centres. Single cases were processed in sub-four hours. Convergence rate was 100% and 75% for single and branched models, respectively. Branching-CA-OCT vFFR had the best agreement with clinical FFR, followed by the single-CA-OCT, followed by the branching-CA models (all suggesting a trend towards superiority to previous state-of-the-art) (Bland Altmann overall bias [limits of agreement]: -0.02 [-0.22 0.18] vs -0.034 [-0.25 0.18] vs -0.001 [-0.24 0.25] respectively). These were statistically non-significant differences.
Conclusions. The novel toolchain and prototype model fused CA-OCT data, generating high-resolution, branched 3D anatomy and simulating clinically useful physiological data. Accuracy was comparable to existing, state-of the art techniques. Further work should focus on reducing processing time, validation in a larger cohort, and improving the graphical user interface.
