Abstract Background Epicardial adipose tissue (EAT) is tightly associated with neighboring myocardium with free diffusion between the two tissues of fatty acid, peptides and adipokines. In specific circumstances EAT can favors myocardial remodeling and, for instance, contribute to the risk of atrial fibrillation (AF). Thereby, there is a growing interest for EAT as a biomarker of AF however, which is limited by difficulties of delineation of EAT vs. other paracardial adipose tissues. The atrio-ventricular groove adipose tissue (GEAT) has been proven to be pure epicardial adipose tissue. Purpose We tested the feasibility and value of measuring pure EAT of the atrio-ventricular groove (GEAT) using routine cardiovascular magnetic resonance imaging (CMR) in a cohort of patients with distinct metabolic disorders and studied its correlates to determine a CMR cardiometabolic score. Methods One hundred patients from the METACARDIS cohort (EU FP7) who had MRI, all in sinus rhythm and without overt clinical heart disease, were divided into 4 groups based on distinct metabolic profiles: metabolic syndrome (MSD, n=25), severe obesity (SO, BMI ≥ 35 kg/m2, n=18), type-2 diabetes (T2D, n=42) and healthy controls (n=15). Atrioventricular groove EAT volume was measured from cine SSFP MRI images in four-, three-, two-chamber and short-axis views; atrial EAT volume was measured from four- and two-chamber views. Native T1-mapping values of EAT at the anterior atrioventricular groove excluding the coronary vessels were measured from the MOLLI sequence (basal short-axis slice). Results CMR was performed in 100 patients from the MetaCardis cohort. GEAT volume measured from end-diastolic long axis views was obtained in all patients with a high correlation between GEAT and atrial EAT (r=0.95; P<0.0001). GEAT volume was higher in the 3 groups of patients with metabolic disorders and highest in MSD. Together with age, body mass, fat distribution, GEAT volume allowed to distinguish obese, T2D and MSD from controls. T2D patients were uniquely characterized in CMR by decreased GEAT T1 relaxation and decreased peak longitudinal left atrial strain. We developed a CMR score by logistic regression and Random Forest machine-learning algorithm combining GEAT volume, T1 and peak LA strain that identified T2D patients with an AUC of 0.81 (Se:77%, Spe:80%; 95%-CI 0.72–0.91, p <0.0001). Conclusion The epicardial adipose tissue of the AV groove measured by CMR can be used as the proxy of atrial EAT and, in combination with atrial strain and T1 relaxation, allows the early identification of the atrial cardiomyopathy associated with diabetes. There is diagnostic value of a multiparametric CMR approach for the detection of subclinical atrial cardiomyopathy and potentially to identify patients at risk of atrial fibrillation and stroke.