Enhanced MACE Risk Stratification Using Automated Machine Learning and PET Imaging

Article Impact Level: HIGH
Data Quality: STRONG
Summary of  Journal of Nuclear Cardiology https://doi.org/10.1016/j.nuclcard.2025.102539 
Dr. Fares Alahdab  et al.

Points

This research study developed an automated machine learning model to predict major adverse cardiovascular events by analyzing clinical data and results from advanced positron emission tomography scans for patients with heart disease.
The artificial intelligence platform demonstrated superior diagnostic performance when compared to traditional logistic regression models and deep feed-forward neural networks in identifying patients at significant risk for future cardiac complications.
Major adverse cardiovascular events tracked during this clinical analysis included patient death as well as myocardial infarction or coronary revascularization procedures occurring more than ninety days after the initial imaging session.
Utilizing automated machine learning allows clinicians to capture complex non-linear relationships between various health factors while providing interpretable patient level explanations that can directly inform and optimize personalized care plans.
Identifying the specific patients who are most at risk for adverse health events is critical for maintaining quality of life and ensuring that medical resources are effectively distributed to those individuals.

Summary

This study evaluated the efficacy of an automated machine learning (AutoML) model in predicting major adverse cardiovascular events (MACE) among patients with suspected or confirmed coronary artery disease (CAD). Traditional risk assessment tools often rely on linear statistical analyses, which struggle to account for complex, non-linear relationships between clinical and imaging variables. The research aimed to determine if AI-driven integration of clinical data and positron emission tomography (PET) imaging could enhance predictive accuracy and provide patient-level explainability.

Researchers utilized data from consecutive patients who underwent clinically indicated PET imaging to train and validate the AutoML model. Predictive performance was compared against a standard logistic regression (LR) model and a deep feed-forward neural network (DNN) using unseen data for testing. MACE was defined as death, myocardial infarction, or coronary revascularization occurring more than 90 days after imaging. The results demonstrated that the AutoML model discriminated MACE risk in CAD more accurately than both the LR and DNN models by effectively capturing non-linear variable interactions.

Beyond increased diagnostic precision, the AutoML framework offers interpretable, case-by-case explanations for its predictions, facilitating personalized clinical decision-making. By identifying patients at high risk for adverse health events through advanced nuclear imaging data, clinicians can better optimize individual care plans and improve long-term quality of life. The study highlights the potential of AutoML as a scalable tool for risk stratification, suggesting its applicability could extend to other complex disease states requiring multi-variable integration.

Link to the article: https://www.sciencedirect.com/science/article/abs/pii/S1071358125004131?via%3Dihub

References

Alahdab, F., El Shawi, R., Ahmed, A. I., Al Rifai, M., & Al-Mallah, M. (2025). Improving prognostic risk assessment of cardiovascular events with machine learning: An evaluation using positron emission tomography myocardial perfusion imaging. Journal of Nuclear Cardiology, 102539. https://doi.org/10.1016/j.nuclcard.2025.102539