Epicardial and paracardial adipose tissue segmentation in short-axis cardiac cine MRI

Motivation: Epicardial and paracardial adipose tissues are important for cardiac disease diagnosis but hard to segment. Goal(s): To develop an approach to segment the two different fat tissues in short-axis cine MRI. Approach: A UNet-based network is proposed to segment the two adipose tissues. Multi-resolution and 3D convolution modules were added to use the motion information for segmentation. Results: The motion information contained in the cine images can substantially help distinguish between the two adipose tissues. Impact: The proposed modules can effectively and precisely segment fat tissues even when the pericardium is challenging to observe, suggesting its potential for clinical applications.