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

Figures

• 

  Figure A1. FCNs DSC performance varies along the cardiac cine frames with a maximum mean Dice and minimum standard deviation in systole.

• 

  Figure A2. FSLeyes plugin interface for epicardial and paracardial segmentation will facilitate reproducibility of proposed methods.

• 

  Figure 1. Overview of the study design.

• 

  Figure 2. Networks’ optimized architecture. The two networks evaluated in this study: U-Net and fully-convolutional network (FCNB) architectures included a first 3D convolution layer to allow multiple cardiac frames as input. Following 2D convolution layers encoded images from 48 features up to 768 features. Eventually, the decoder targeted three labels for segmentation in the central input frame: epicardial adipose tissue (EAT), paracardial adipose tissue (PAT), and heart ventricles (HV).

• 

  Figure 3. Comparison of reference total epicardial fat volume andproposed EAT area measured on four-chamber cine. EAT area was measured in end-systolic or end-diastolic frame across the 100 subjects’ database. The three cohorts merged for the database were identified by markers color.

• 

  Figure 4. Representative segmentation results for each population defined by quartile of EAT area. Images were cropped around the heart for visualization. White arrows point out discrepancies between manual and automatic segmentations. As detailed in the methods, only periventricular EAT was segmented.

• 

  Figure 5. Quartile classification results from EAT area estimated from networks segmentation against classification from manual EAT area. Only segmentations from preferred systolic frames were shown here. Markers colors were defined by manual EAT quartiles. Red squares delineate manual EAT quartiles.