Rapid cardiac MR myocardial perfusion quantification using machine learning trained with synthetically generated sample data

摘要

While myocardial hypoperfusion is routinely diagnosed by visual assessment of dynamic contrast-enhanced (DCE) cardiac MRI, exact quantification of perfusion parameters (MBF = myocardial blood flow, PS = permeability surface area product, Vp = plasma volume, Ve = extracellular volume) is desirable. The blood tissue exchange model (BTEX)1 applied in recent studies2 offers detailed modelling, but its complexity increases computational costs and vulnerability to noise when applying conventional fitting. Our study sought to predict perfusion parameters fast and accurately using a convolutional neural network (CNN) trained with synthetically generated sample data.