In this study, we have implemented and evaluated a new 3D cine Dixon technique for the quantification of epicardial adipose tissue. Compared to standard single-phase Dixon, the proposed method yields fat quantification values with lower inter- and intraobserver variability.
The availability of images throughout the cardiac cycle allows retrospective selection of a phase with potentially better visualization of the epicardial fat border. Conversely, for the standard single-phase approach, the trigger delay is prospectively determined and the timing for optimal epicardial fat border detection is unknown. Examples of this have been provided (Fig. 2) where the single-phase Dixon acquired in the phase with longest rest period yield images with suboptimal delineation of the epicardial fat depot, while better visualization is obtained with cine Dixon in the opposite (systolic/diastolic) cardiac rest period. A corollary to this is that no specific timing information is required prior to the cine Dixon scan thus simplifying scan setup, while additional time-resolved 2D scans are needed to prospectively define the single-phase Dixon timing parameters (21).
The ability to visualize the motion of the epicardial fat may also provide information to help separate epicardial from paracardial fat, even if the border between the tissues is inconspicuous. This may be particularly beneficial for identifying the border of the epicardial fat near the atrioventricular groove and the basal part of the right ventricle, which typically experience significant longitudinal motion during the cardiac cycle relative to the more static adjacent paracardial fat. An example of this motion-related differentiation of epicardial and paracardial fat have been provided (Fig. 3 and supplementary video), and this advantage of cine Dixon may further help to explain why segmentation is less variable using this approach compared to single-phase Dixon.
In this study, we found a lower intra-and inter-observer variability using a novel cine Dixon technique despite slightly reduced image quality of the water image compared to conventional single-phase Dixon. Although there was no significant difference for the fat images, segmentation was performed using the fat fraction images which combines the water and fat images. However, the small reduction in image quality using cine Dixon is likely primarily due to residual motion artifacts. A diaphragmatic navigator was used to gate the cine acquisition to end-expiration and performed once per cardiac cycle, resulting in low temporal proximity to the image acquisition in many frames. However, both navigator gating and slice tracking was employed for single-phase Dixon, where the navigator was performed immediately prior to the data readout for each cardiac cycle. Recent improvements in respiratory motion compensation techniques for whole-heart cine such as self-navigation are likely to prove beneficial in combination with cine Dixon for epicardial border delineation and can be included in future work to further improve this technique (22–29).
The assessed epicardial fat volume was systematically higher using the single-phase Dixon sequence compared to the proposed cine Dixon sequence. It is reasonable to believe that the epicardial fat segmentation for the single-phase Dixon could be overestimated due to the inclusion of the adjacent paracardial fat if the border between these two adipose tissue depots is diffuse or blurred in that particular cardiac phase. Conversely, for the cine-Dixon, the border may be more precisely inferred by interpolating from alternative time frames where it may be more clearly visible, as shown in Figs. 2 and 3 and supplementary video.
Body mass index is a common anthropometric index of overweight and obesity. Interestingly, epicardial fat volume index to BSA showed a significant relation to BMI using the proposed cine Dixon technique but not using the single-phase technique. However, the relation was only modest implying that BMI is a rather unspecific metric of ectopic adipose tissue (30). In previous studies using CMR Dixon imaging of non-cardiac adipose tissue depots, not only the fat volume but also the fat concentration has been assessed (31). This information has not yet been obtained for cardiac adipose tissue but would likely provide additive value in terms risk assessment of cardiovascular disease.
Compared to previous work using single-phase Dixon for fat quantification by Homsi et al. (14), we have obtained a similar mean measurement error (bias) for the inter-observer variability comparison (0.42 ml per slice × 10 slices per patient = 4.2 ml per patient in this study vs. 4.5 ml in Homsi et al.) but with a higher standard deviation (1.53 ml per slice in this study, 15.3 ml per patient vs. 4.1 ml in Homsi et al.). This may be due to different acquisition protocols, where the images in this study were acquired after contrast injection while in Homsi et al. the Dixon images were acquired without contrast.
The implemented fat quantification methods rely on the fat fraction variable to classify voxels as fat (14). The advantage of using the fat fraction for quantification is that flow-related artifacts which may otherwise be classified as fat are suppressed and simple threshold-based segmentation techniques may be readily applied. Compared to previous work which use a subject-specific threshold obtained by segmenting areas of pure fat and myocardium (14), we used a fixed threshold for all datasets. This minimizes variability in quantification due to differences in segmentation of the fat and myocardium region-of-interests which may confound the variability of segmentation of the epicardial fat. Due to the low steady-state signal of water signal for the spoiled gradient recalled-echo sequence used here, a T1-shortening contrast agent was necessary to facilitate the fat quantification for the cine Dixon technique. However, the fat signal is unaffected by the contrast agent and the use of a contrast agent is therefore not a strict requirement, its primary purpose is to support the classification problem. Classification methods that only utilize the fat signal can be developed which relaxes this requirement.
A potential bias with regard to the contrast agent administration could be caused by the non-randomization of scan order (cine Dixon was always performed first to maximise contrast agent effect, while the single phase scan was performed second as it does not benefit from contrast agents to the same extent). However, the contrast agent (gadobutrol) has a relatively long washout period and a half-life of approximately 2 hours (32). The 5 to 7-minute difference in time after contrast agent injection between cine and single-phase Dixon therefore is unlikely to significantly affect the outcomes.
A limitation of the cine Dixon technique compared to single-phase Dixon is the longer scan time which is due to the higher temporal resolution of cine Dixon. In this study we used a temporal resolution of 65 ms for the cine Dixon scan, while the single-phase resolution was approximately 100 ms, leading to a proportional increase in scan time. The lower temporal resolution of the single-phase Dixon is unlikely to significantly reduce image quality as it was adapted to coincide with the patient-specific rest period. Maintaining a high temporal resolution for cine Dixon is important to mitigate cardiac motion blurring and capitalize on the captured cardiac motion between frames which facilitates delineation of the epicardial fat border. However, by employing compressed sensing with high image acceleration factors a nominal scan time of 2 minutes and 45 seconds was achieved, which was considered clinically acceptable.