Direct Risk Assessment From Myocardial Perfusion Imaging Using Explainable Deep Learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Myocardial perfusion imaging (MPI) is crucial for risk stratification in cardiac patients.
• Current MPI methods require improvement in predictive accuracy.
• Accurate prediction of adverse cardiac events is essential for patient management.

Purpose of the Study:

• To develop and validate an explainable deep learning (DL) model for predicting major adverse cardiac events (MACE).
• To assess the HARD MACE-DL model's performance against traditional quantitative approaches.
• To enhance the accuracy and interpretability of risk prediction using MPI data.

Main Methods:

• Development of the HARD MACE-DL model using myocardial perfusion, motion, thickening, phase polar maps, and clinical data.
• Inclusion of 20,401 patients for training/internal testing and 9,019 for external testing.
• Evaluation of prognostic accuracy using the area under the receiver-operating characteristic curve (AUC) and calibration metrics.

Main Results:

• HARD MACE-DL demonstrated superior accuracy (AUC 0.73) compared to logistic regression (AUC 0.70), stress TPD (AUC 0.65), and ischemic TPD (AUC 0.63) in external testing.
• Patients with high HARD MACE-DL scores showed a 10-fold higher annual risk of death or MI compared to those with low scores.
• Excellent calibration was observed in both internal and external testing groups.

Conclusions:

• The explainable DL model (HARD MACE-DL) accurately predicts death or MI directly from MPI data.
• This model offers improved prognostic accuracy and calibration over traditional methods.
• The integrated explainability feature allows physicians to identify image regions contributing to risk predictions.