AI-Enabled CT Cardiac Chamber Volumetry Predicts Atrial Fibrillation and Stroke Comparable to MRI.
Summary
In 3,552 MESA participants followed for 15 years, AI-derived LA volume from CAC CT predicted incident AF and stroke comparably to CMRI LA volume (AUC ~0.80 for AF; ~0.76 for stroke) and improved 5-year AF risk reclassification when added to CHARGE-AF, NT-proBNP, and Agatston scores.
Key Findings
- AI-derived LA volume from CAC CT predicted incident AF and stroke with AUC similar to CMRI-derived LA volume (AF: 0.802 vs 0.798; stroke: 0.762 vs 0.751).
- Adding AI-CAC LA to CHARGE-AF, NT-proBNP, and Agatston scores significantly improved 5-year AF risk reclassification (positive continuous NRI).
- Long-term (15-year) outcomes confirm robustness of AI volumetry across asymptomatic, multi-ethnic cohort.
Clinical Implications
Radiology pipelines could incorporate AI LA volumetry on CAC scans to flag high AF/stroke risk, informing preventive strategies and closer rhythm monitoring without new imaging.
Why It Matters
Enables opportunistic AF/stroke risk prediction from widely available CAC CT without dedicated CMRI, leveraging AI to extract additional prognostic value.
Limitations
- Observational design limits causal inference; external clinical utility and workflow integration require prospective implementation studies.
- Generalizability to symptomatic populations and scanner/protocol variability needs assessment.
Future Directions
Prospective clinical deployment to test triggered monitoring/prevention strategies, calibration across scanners/vendors, and cost-effectiveness analyses.
Study Information
- Study Type
- Cohort
- Research Domain
- Diagnosis/Prognosis
- Evidence Level
- II - Prospective cohort analysis with long-term outcomes.
- Study Design
- OTHER