Machine learning applications in cardiac computed tomography: a composite systematic review.

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence

Background:

• Cardiac computed tomography (CT) is increasingly analyzed using artificial intelligence (AI) and machine learning (ML).
• Significant advancements have been made in integrating ML with cardiac CT for various diagnostic applications.

Purpose of the Study:

• To provide an overview of contemporary advances in cardiac CT analysis driven by ML.
• To review the latest studies and evolving techniques combining ML and cardiac CT.

Main Methods:

• A systematic literature search was conducted across Medline, Embase, and Cochrane Library up to November 2021.
• Studies focused on six key areas: CT-fractional flow reserve (CT-FFR), atrial fibrillation (AF), aortic stenosis, plaque characterization, fat quantification, and coronary artery calcium scoring.
• 57 relevant studies were included in the review.

Main Results:

• ML algorithms accurately estimate non-invasive CT-FFR, potentially reducing the need for invasive angiography.
• Automated and accurate calculation of coronary artery calcification and non-calcified lesions is now possible.
• ML enables rapid and accurate quantification of epicardial adipose tissue.
• Streamlined ML algorithms improve aortic annular measurements for transcatheter aortic valve replacement (TAVR) planning.
• ML-driven segmentation of the left atrium (LA) aids in predicting post-ablation atrial fibrillation recurrence.

Conclusions:

• ML significantly enhances the capabilities of cardiac CT, offering more accurate and efficient diagnostic tools.
• These AI-driven advancements have the potential to improve patient outcomes in cardiovascular care, from diagnosis to treatment planning.