Deep learning-based quantification of epicardial adipose tissue volume in cardiac magnetic resonace predicts major adverse cardiovascular events in patients with known or suspected CAD

Abstract Funding Acknowledgements Type of funding sources: None. Background Increased epicardial adipose tissue (EAT) volume quantified with cardiac magnetic resonance (CMR) has been associated with the development of major adverse cardiac events (MACE). We sought to investigate the additional prognostic role EAT volume in patients with known or suspected coronary artery disease (CAD) undergoing CMR imaging. Methods 702 consecutive patients (age: 63±10 y, male 84%) with known or suspected CAD underwent clinically indicated CMR. Using a new DL algorithm, EAT volume was quantified on short-axis stack steady state free precession (SSFP) images. Firstly, a training set of 300 patients with manually traced EAT (ground truth) was randomly split (patient-wise) into development (n=240, 80%) and held-out testing (n=60, 20%) cohorts. Secondly, we applied our segmentation network on a validation set of 402 patients with unlabeled data for automated EAT segmentations. Finally, we applied the algorith to the entire population (702 patients) to quantify EAT. EAT volume, normalized for the body mass index (EAT index), was compared to standard clinical and imaging variables for the prediction of MACE defined as non-fatal myocardial infarction and cardiac deaths. Results 52 patients (7.4%) developed MACE during a follow-up of 5.8±1.2 years. Left ventricular ejection fraction (LVEF) < 50% (HR 2.271 [95% CI 1.117–4.616]), p = 0.023, late gadolinium enhancement (LGE) presence (HR 2.456 [95% CI 1.077–5.602]), p = 0.033 and EAT index ≥ 1,8 (HR 6.187 [95% CI 1.879–20.372]), p = 0.003 were independent predictors of MACE. Adding EAT index in a model including LVEF and LGE provided a significant improvement in predicting the endpoint with a Harrell C statistic of 0.75. Conclusions In patients with known or suspected CAD undergoing CMR, fully automated EAT volume quantification provides additional prognostic information on top of standard clinical and imaging parameters.