A personalized computational framework for the diagnosis of cardiac perfusion defects

Background and objectives: Myocardial Blood Flow (MBF) is a key indicator of myocardial perfusion, typically assessed through additional clinical tests like dynamic CT perfusion under stress. This study introduces a computational framework designed to enhance coronary artery disease diagnosis by predicting MBF using data from routine CT images and clinical measurements. Methods: The computational framework employs AI methods to reconstruct coronary and myocardial geometries and integrates a computational model, featuring 3D coronary arteries and a three-compartment myocardial model, blindly calibrated with data from six representative patients. Results: Validation on 28 additional patients showed MBF predictions consistent with experimental and clinical measurements. Confusion matrix analysis assessed the twin's ability to classify pathological (averaged MBF < 240 ml/min/100 g) versus healthy perfusion regions, yielding a recall of 0.81, with precision of 0.68 and accuracy at 0.7. Conclusions: This work represents the first attempt to predict and validate MBF on such a large cohort, paving the way for future clinical applications.