Integrated AI workflow for non-invasive coronary intervention

Invasive coronary angiography is performed on around 100,000 patients a year in the UK and is used to decide their trajectory of revascularisation, be it e.g. percutaneous intervention or surgical bypass graft. The guidance of the procedures is achieved exclusively via two-dimensional x-ray fluoroscopic imaging, which may be difficult to interpret in complex cases and also hinders the adoption of standardised quantitative metrics in diagnosis. We have previously developed a proof-of-concept AI method to reconstruct 3D vascular network anatomy from the routine 2D angiograms. Its output can also serve as a starting point for estimating pressure gradients, eliminating the need for invasive measurements. Building on the above, this project will finetune the reconstruction method on patient data, develop vessel phantoms for validation, and investigate non-invasive assessment of vascular pressures using physics-informed neural networks, for a fully functional clinical prototype of integrated coronary assessment to be used in cathlab examinations.