Personalized Cardiac Computer Models and Machine-Learning for Target Vein Selection in Cardiac Resynchronisation Therapy

Cardiac resynchronization therapy (CRT) is the effective treatment of heart failure with a reduced ejection fraction. However, in up to 30% of patients, the left ventricular ejection fraction (LV EF) does not improve after CRT device implantation. The selection of the candidates for CRT, device implantation planning, and optimization of pacing electrode position remain vital clinical challenges. In our previous studies, we developed a new technology that predicts CRT outcomes. The technology combines clinical information, computer simulation, and machine learning (ML). Our approach calculates a ML score ranging from 0 to 1 that classifies a positive or negative prediction for LV EF improvement depending on the epicardial position of LV pacing site during biventricular (BiV) pacing and identifies an optimal LV pacing site that maximizes the ML-score for a patient. In this work we have essentially extended our previous pipeline, adding detailed segmentation of myocardial fibrosis and coronary veins to our model simulations. We tested the extended pipeline using clinical data from 19 patients with severe fibrosis, where 13 (68%) patients did not improve LV EF during one year of BiV pacing. For this group of patients, the specificity of classifier of CRT response reached 1. For the majority of non-responders (11 (85%) of 13) from the group, our technique does not predict the possibility of improvement by LV pacing site optimisation using coronary venous system. However, in three patients, our technology predicted a positive response to CRT if the target LV lead position had been optimised prior to implantation. The results of this study suggest the potential of our approach for pre-implant patient selection and LV pacing site optimisation in selected patients.