Related Experiment Video
Updated: Jun 9, 2025
Author Spotlight: Decellularization-Based Quantification of Skeletal Muscle Fatty Infiltration
Published on: June 9, 2023
An enhanced deep learning method for the quantification of epicardial adipose tissue
Ke-Xin Tang1,2, Xiao-Bo Liao3, Ling-Qing Yuan2
1Department of Radiology, the Second Xiangya Hospital, Central South University, No. 139 Middle Renmin Road, Furong District, Changsha, 410000, China.
This study introduces an enhanced deep learning method for accurately quantifying epicardial adipose tissue (EAT) volume from coronary computed tomography angiography (CCTA) scans. The improved technique integrates anatomical information, achieving high agreement with manual measurements for cardiovascular disease risk assessment.
Area of Science:
- Cardiology
- Medical Imaging
- Artificial Intelligence
Background:
- Epicardial adipose tissue (EAT) volume is a significant contributor to cardiovascular disease (CVD) progression.
- Manual quantification of EAT is time-consuming and prone to errors.
- Existing deep learning methods for EAT quantification often lack interpretability and anatomical context.
Purpose of the Study:
- To develop and validate an enhanced deep learning method for automatic EAT volume quantification using coronary computed tomography angiography (CCTA).
- To improve the accuracy and interpretability of EAT quantification by integrating morphological information.
Main Methods:
- A novel deep learning approach was developed, incorporating both data-driven techniques and specific pericardial morphological information.
- The method quantified EAT volume based on CT attenuation values within the predicted pericardium.
- The study included 108 patients, with data randomly assigned to training (n=60), validation (n=8), and test (n=40) sets.
Main Results:
- The enhanced deep learning method demonstrated strong agreement with expert manual quantification.
- Median Dice score coefficients (DSC) were 0.916 for 2D slices and 0.896 for 3D volumes.
- An excellent correlation (0.980, p<0.001) and low bias (-2.39 cm³) were observed for EAT volume measurements.
Conclusions:
- Integrating pericardial anatomical structures enhances deep learning-based automatic EAT quantification.
- The proposed method shows high accuracy and reliability for EAT volume assessment.
- The findings suggest significant potential for clinical application in CVD risk stratification.
