Fully Automated CT Quantification of Epicardial Adipose Tissue by Deep Learning: A Multicenter Study

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Epicardial adipose tissue (EAT) is an independent predictor of cardiovascular events.
• Accurate quantification of EAT is crucial for cardiovascular risk assessment.
• Current manual quantification methods are time-consuming and prone to interobserver variability.

Purpose of the Study:

• To evaluate the performance of a deep learning algorithm for automated EAT quantification.
• To assess the robustness and accuracy of deep learning compared to expert readers.
• To determine the feasibility of implementing deep learning for routine clinical use.

Main Methods:

• A convolutional neural network was trained on 850 multicenter cardiac CT scans.
• Deep learning performance was compared with three expert readers and interobserver variability.
• Automated EAT progression was analyzed in patients with baseline and follow-up scans.

Main Results:

• Deep learning automated EAT quantification in 1.57 seconds, significantly faster than experts (15 minutes).
• High agreement was observed between deep learning and expert EAT quantification (R = 0.974).
• Deep learning accurately tracked EAT progression, correlating strongly with manual measurements (R = 0.905).

Conclusions:

• Deep learning enables rapid, robust, and fully automated EAT quantification from cardiac CT.
• The algorithm performs comparably to expert readers.
• This technology can be implemented for routine cardiovascular risk assessment.