Deep learning applications in coronary anatomy imaging: a systematic review and meta-analysis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Deep learning (DL) applications in medical imaging are increasingly prevalent.
• Coronary artery disease (CAD) is a primary focus for DL in imaging.
• Accurate imaging of coronary artery anatomy is crucial for CAD diagnosis and management.

Purpose of the Study:

• To systematically review the evidence on the accuracy of DL applications in coronary anatomy imaging.
• To assess the performance of DL models in various CAD imaging tasks.
• To evaluate the translation of DL technologies into clinical practice for CAD.

Main Methods:

• Systematic literature search on MEDLINE and EMBASE databases.
• Inclusion of studies applying DL to coronary anatomy imaging.
• Meta-analysis of studies predicting fractional flow reserve (FFR) using coronary computed tomography angiography (CCTA).
• Quality assessment using QUADAS and heterogeneity testing (tau², I², Q tests).

Main Results:

• 81 studies met inclusion criteria; CCTA (58%) and CNNs (52%) were most common.
• Most studies reported good performance, with Area Under the Curve (AUC) ≥80% for various tasks.
• Pooled diagnostic odds ratio (DOR) for FFR prediction was 12.5 (8 studies); no significant heterogeneity.

Conclusions:

• DL, especially CNNs, demonstrates powerful performance in coronary anatomy imaging.
• Many DL applications require external validation before widespread clinical adoption.
• Emerging applications like CT-FFR are translating into clinical practice, potentially improving CAD patient care.