An enhanced deep learning method for the quantification of epicardial adipose tissue

Figures

• 

  Figure 1. Labeled images during the labeling process. The labeling process is implemented in the order from (A) to (D). Part (A) is the original contrast CT slice, (B) is the drawing of pericardium, (C) is the label of the pericardium and its inner region in axial, sagittal, coronal and 3D views, (D) is the EAT fat pixels in Axial and 3D views.

• 

  Figure 2. The architecture of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{\theta\:}$$\end{document}. Each green and orange item represents a multi-channel feature map. The number of channels is shown at the top of the volume, and the length and width are provided at the lower-left edge of the volume. The arrows denote the different operations, which are explained at the lower-right corner of the figure.

• 

  Figure 3. An illustration for connected components in 2D case and 1D case. The white area A, B, C, D in the 2D case denote different connected components in the binary image. Similarly, the white area a, b, c, d in the 1D case denote different connected components in the binary sequence.

• 

  Figure 4. The three steps of our post-processing method.

• 

  Figure 5. Expert 1 vs. expert 2 EAT quantifications. A high ICC is obtained (ICC = 0.993, p < 0.001).

• 

  Figure 6. Segmentation results of our algorithm. Each row presents the segmentation results of MIDL and the expert for three contrast CT slices. Top, middle, and bottom rows correspond to superior, median, and inferior region of the heart, respectively. The red region indicates the EAT.

• 

  Figure 7. MIDL vs. expert EAT quantifications. An excellent agreement is obtained with high correlation (R = 0.980, p < 0.001).

• 

  Figure 8. Bland-Altman analysis of MIDL and expert EAT quantifications. The analysis demonstrates a non-significant bias between the two measures (-2.39 cm3, p = 0.089).

• 

  Figure 9. 2D segmentation results of different methods. Each row presents the EAT segmentation results of three algorithms and the expert segmentation for one contrast CT slice. The red part indicates the EAT.

• 

  Figure 10. 3D segmentation results of different methods. Each row presents the 3D EAT segmentation results from three algorithms along with the expert segmentation for the CCTA series. Each column displays two different views of the 3D EAT from two patients.

• 

  Figure 11. The framework of our proposed deep learning method.