Holistic AI analysis of hybrid cardiac perfusion images for mortality prediction.

Summary

In 10,480 patients across four centers, a holistic AI model that combined CT attenuation correction radiomics, calcium, epicardial fat, perfusion, stress test, and clinical features achieved an AUC of 0.80 for all-cause mortality, outperforming calcium scoring or perfusion alone. Multi-organ, multi-feature integration significantly improved prognostic performance.

Key Findings

• Multi-structure AI segmentation and radiomics from CTAC across 33 organs improved risk prediction.
• Integrated model (MPI + CT + stress + clinical) achieved AUC 0.80 for all-cause mortality.
• Model outperformed coronary calcium scoring (AUC 0.64) and perfusion alone (AUC 0.62), p<0.001.

Clinical Implications

Hybrid MPI protocols could incorporate automated AI extraction of CTAC radiomics and cardiac adiposity to provide decision-ready mortality risk scores, complementing traditional perfusion metrics.

Limitations

• Retrospective design; potential confounding and dataset shift
• Clinical utility thresholds and prospective impact on decision-making not tested

Future Directions

Prospective, multi-center impact studies embedding AI risk scores into clinical workflows; exploration of causal links between specific CTAC radiomics and pathophysiology.