Automatic epicardial adipose tissue segmentation in pulmonary computed tomography venography using nnU-Net
Yifan Hu1, Shanshan Jiang2, Xiaojin Yu1
1Department of Radiology, Dongtai People's Hospital, Yancheng, China.
Quantitative Imaging in Medicine and Surgery
|October 23, 2023
Summary
A deep learning model automates epicardial adipose tissue (EAT) quantification from pulmonary computed tomography venography (PCTV) scans. This AI approach significantly reduces analysis time and shows high consistency with manual measurements.
Area of Science:
- Cardiology
- Medical Imaging
- Artificial Intelligence
Background:
- Epicardial adipose tissue (EAT) plays a crucial role in cardiac pathophysiology.
- Investigating EAT is vital for understanding heart disease mechanisms.
Purpose of the Study:
- To develop a deep learning (DL) model for automated EAT extraction and quantification.
- To assess the efficacy of DL in analyzing EAT from pulmonary computed tomography venography (PCTV) images.
Main Methods:
- A DL model was trained and validated on PCTV datasets from 128 patients (including internal and external test sets).
- Automated EAT segmentation and quantification were performed using the DL model.
- Results were compared against manual quantification using Dice score coefficient (DSC), Hausdorff distance (HD95), and correlation analyses.
Main Results:
- The DL model achieved successful automated EAT segmentation in all cases across internal and external test sets.
- Analysis time was significantly reduced with DL (5.43±2.52 min) compared to manual methods (106.20±15.90 min).
- High consistency (DSC: 0.92±0.02 internal, 0.88±0.03 external) and excellent correlation (r > 0.9) were observed between DL and manual quantification.
Conclusions:
- The developed DL model enables fully automatic and accurate quantification of EAT from PCTV images.
- The DL model's performance is highly consistent with traditional manual quantification methods, offering a more efficient alternative.
