Deep-Learning Segmentation of Epicardial Adipose Tissue Using Four-Chamber Cardiac Magnetic Resonance Imaging

Abstract

In magnetic resonance imaging (MRI), epicardial adipose tissue (EAT) overload remains often overlooked due to tedious manual contouring in images. Automated four-chamber EAT area quantification was proposed, leveraging deep-learning segmentation using multi-frame fully convolutional networks (FCN). The investigation involved 100 subjects—comprising healthy, obese, and diabetic patients—who underwent 3T cardiac cine MRI, optimized U-Net and FCN (noted FCNB) were trained on three consecutive cine frames for segmentation of central frame using dice loss. Networks were trained using 4-fold cross-validation (n = 80) and evaluated on an independent dataset (n = 20). Segmentation performances were compared to inter-intra observer bias with dice (DSC) and relative surface error (RSE). Both systole and diastole four-chamber area were correlated with total EAT volume (r = 0.77 and 0.74 respectively). Networks’ performances were equivalent to inter-observers’ bias (EAT: DSCInter = 0.76, DSCU-Net = 0.77, DSCFCNB = 0.76). U-net outperformed (p < 0.0001) FCNB on all metrics. Eventually, proposed multi-frame U-Net provided automated EAT area quantification with a 14.2% precision for the clinically relevant upper three quarters of EAT area range, scaling patients’ risk of EAT overload with 70% accuracy. Exploiting multi-frame U-Net in standard cine provided automated EAT quantification over a wide range of EAT quantities. The method is made available to the community through a FSLeyes plugin.

Motion‐resolved fat‐fraction mapping with whole‐heart free‐running multiecho GRE and pilot tone

“…The free‐running ME‐GRE framework enabled quantitative fat measurements across the cardiac cycle, which allows to select a preferred frame for various adipose tissue characterization, 67 and in the current volunteer study, a variable FF pattern was consistently observed as a function of time, indicating an heterogeneity of the pericardial fatty tissue. This result suggests that motion‐averaged or frozen‐motion (i.e., triggered) visualizations might only allow for partial or incomplete tissue characterization, a finding that can be linked to similar reports on the effect of respiratory motion on R 2 * quantification in the liver 68,69 .…”

Role of epicardial adipose tissue in diabetic cardiomyopathy through the lens of cardiovascular magnetic resonance imaging – a narrative review

“…Two studies presented novel MRI techniques to assess EAT quantification. Daudé et al 53 employed fully convolutional networks that were used for cardiac volume segmentation techniques to measure the EAT area on horizontal long-axis four-chamber cine MRI multi-frame images. This approach was developed to overcome interobserver variability and shorten the time-consuming nature of manual analysis.…”

“…This approach was developed to overcome interobserver variability and shorten the time-consuming nature of manual analysis. 53 …”

“…To overcome the time-consuming nature of manual EAT segmentation, semiautomated and fully automated artificial intelligence (AI)-based programmes have been developed. 53 , 54 Currently used manual methods to contour EAT may take up to 13 min by an expert cardiologist, whereas this is reduced to 5 min by automated methods. 55 …”

The Role and Implications of Epicardial Fat in Coronary Atherosclerotic Disease

“…Transthoracic echocardiography is the main tool used to quantify and measure the thickness and the volume of epicardial fat [ 39 ]. Computer tomography (CT) and magnetic resonance imaging (MRI) are mainly used to evaluate the volume of EAT because of their high spatial resolution [ 40 ].…”