A novel approach for the automated segmentation and volume quantification of cardiac fats on computed tomography

Area of Science:

• Medical Imaging
• Cardiovascular Health
• Artificial Intelligence in Medicine

Background:

• Cardiac fat deposits, including epicardial and mediastinal fat, are linked to various cardiovascular diseases like atherosclerosis and atrial fibrillation.
• These fat deposits are independent of overall obesity, highlighting the need for precise quantification.
• Manual segmentation of cardiac fats is labor-intensive and costly, limiting its clinical application.

Purpose of the Study:

• To develop a unified, autonomous method for segmenting and quantifying epicardial and mediastinal cardiac fats.
• To minimize user intervention in the cardiac fat segmentation process.
• To compare the efficacy of different classification algorithms for this task.

Main Methods:

• The proposed methodology integrates registration and classification algorithms for autonomous segmentation.
• Various classification algorithms, including neural networks, probabilistic models, and decision trees, were evaluated.
• The focus was on achieving minimal user interaction throughout the process.

Main Results:

• The autonomous method achieved a mean accuracy of 98.5% for both epicardial and mediastinal fats (99.5% with feature normalization).
• A mean true positive rate of 98.0% was recorded.
• The average Dice similarity index reached 97.6%, indicating high segmentation precision.

Conclusions:

• The developed unified method effectively automates cardiac fat segmentation and quantification.
• The high accuracy and efficiency of the proposed method offer a valuable tool for clinical practice.
• This automated approach has the potential to reduce healthcare costs and improve cardiovascular risk assessment.