AI-based detection and classification of anomalous aortic origin of coronary arteries using coronary CT angiography images.

Summary

A fully automated deep learning pipeline achieved near-perfect performance (AUC ≥0.99; sensitivity/specificity 0.95–0.99) for detecting and classifying AAOCA on 3D-CCTA across internal and external datasets. The system supports real-time alerts and scalable cohort analyses, enabling clinical integration and population-level screening studies.

Key Findings

• AI achieved AUC ≥0.99 with sensitivity and specificity 0.95–0.99 across internal and external datasets.
• End-to-end, fully automated pipeline supports real-time alerts for high-risk AAOCA anatomies.
• Tool enables large-scale 3D-CCTA cohort analyses to refine epidemiology and risk stratification.

Clinical Implications

Integration into CCTA reading could enable automated triage and consistent detection of high-risk AAOCA variants, prompting further anatomical assessment and risk counseling, especially in athletic or military screening.

Limitations

• Retrospective datasets; prospective clinical impact studies are needed
• Potential class imbalance and heterogeneity across scanners/sites not fully detailed

Future Directions

Prospective, multicenter trials to test workflow integration, impact on diagnostic accuracy and outcomes; regulatory pathways for clinical deployment; expand to other congenital anomalies.