MRI Protocol and Image Analysis

At the time of study entry, ECG‐gated 1.5‐T MRI (Philips Achieva 1.5 Tesla; Philips Healthcare, Best, the Netherlands) was performed with a dedicated cardiac coil for volumetric quantification of cardiac adipose tissue, including both epicardial and pericardial fat. The body coil was used for volumetric quantification of abdominal subcutaneous adipose tissue (SAT) and VAT as well as to estimate hepatic fat fraction. Axial T1‐weighted (T1W) spin echo sequence (slice thickness=5 mm; repetition time [TR]=750 ms; echo time [TE]=6–8 ms; resolution=0.67×0.67 mm) was performed at the level of the mitral valve to quantify cardiac fat during diastole. Axial T1W slice was also acquired at the level of L4‐L5 intervertebral space for VAT and SAT quantification (slice thickness=5 mm; TR=750 ms; TE=6–8 ms; resolution=0.78×0.78 mm).6 For hepatic fat measurement, axial T1W slice was acquired at the level of L1‐L2 lumbar intervertebral space (slice thickness=5 mm; TR=750 ms; TE=6–8 ms; resolution=0.78×0.78 mm). All acquisitions were performed with and then without fat saturation to confirm that the disappearing signal is in fact fat.

Image analysis was performed off‐line in a standardized core laboratory (Laboratoire d'Imagerie Cardiovasculaire Avancée) using dedicated software (QMass MR; Medis, Leiden, the Netherlands) by trained technicians blinded to the study hypothesis and patient data. On fat‐enhanced images, the pericardial space was identified as the dark stripe separating the bright epicardial fat within from the pericardial fat without, and thus cardiac fat was readily traced manually. On fat‐enhanced images, the SAT contour was well delineated and manually traced. To minimize observer bias and maximize reproducibility of VAT quantification, a standardized region of interest was positioned in the homogeneous adipose tissue, and all pixels with similar signal intensities were automatically classified as VAT by the software. Cardiac and abdominal fat volumes were reported for a standardized length of 5 mm (mm³/5 mm). Hepatic fat fraction was calculated as the difference between the signal from the non–fat‐saturated image and the signal from the fat‐saturated image/the signal from the non–fat‐saturated image, and reported as a percentage. Detailed method is available in a prior publication.6