Cardiac fat segmentation using computed tomography and an image-to-image conditional generative adversarial neural...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Increased cardiac adipose tissue is linked to cardiovascular diseases like atrial fibrillation and coronary heart disease.
• Manual segmentation of cardiac fat is time-consuming and costly, limiting clinical application.
• There is a need for efficient and accurate computational methods for cardiac fat quantification.

Purpose of the Study:

• To develop and evaluate a novel deep learning methodology for autonomous segmentation and quantification of epicardial and mediastinal fats.
• To assess the efficacy of the pix2pix network for cardiac fat segmentation.
• To provide a precise and time-efficient alternative to manual segmentation methods.

Main Methods:

• A deep learning approach utilizing the pix2pix generative adversarial network architecture was employed.
• The methodology focused on the autonomous segmentation of two distinct cardiac fat types: epicardial and mediastinal fats.
• Performance was evaluated based on accuracy and f1-score for segmentation precision.

Main Results:

• The deep learning model achieved high accuracy and f1-scores for both fat types: 99.08% accuracy and 98.73 f1-score for epicardial fat, and 97.90% accuracy and 98.40 f1-score for mediastinal fat.
• The proposed method demonstrated superior performance compared to existing studies in terms of f1-score and processing time.
• Real-time image segmentation was achieved, significantly reducing analysis duration.

Conclusions:

• The novel deep learning methodology provides highly accurate and efficient autonomous segmentation of cardiac fat deposits.
• This approach holds significant potential for improving clinical risk assessment of cardiovascular diseases.
• The pix2pix network proves effective for cardiac fat segmentation, offering a valuable tool for medical professionals.