Artificial intelligence based automatic quantification of epicardial adipose tissue suitable for large scale...

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Epicardial adipose tissue (EAT) is linked to cardiometabolic risk.
• Accurate and large-scale quantification of EAT is challenging.
• Existing methods for EAT measurement can be time-consuming and prone to error.

Purpose of the Study:

• To develop a fully automatic model for quantifying epicardial adipose tissue (EAT) volumes and attenuation.
• To enable reliable EAT assessment in large population studies.
• To investigate the relationship between EAT and cardiometabolic risk markers.

Main Methods:

• Utilized a convolutional neural network (CNN) trained on non-contrast cardiac CT images from the SCAPIS study.
• Implemented automatic segmentation of the pericardium and artifact suppression.
• Developed an imputation method for missing EAT volumes in incomplete image sets.

Main Results:

• The model achieved a high Dice coefficient of 0.90 against expert segmentations.
• Successfully segmented 99.4% of 1400 tested image sets.
• Automatic imputation of missing EAT volumes showed an error of less than 3.1%.

Conclusions:

• An effective, automated model for large-scale EAT quantification has been developed.
• The model addresses common challenges in EAT measurement from cardiac CT.
• Future studies should consider the strong correlation between EAT and anthropometric measures.