Interpretable Prediction of Myocardial Infarction Using Explainable Boosting Machines: A Biomarker-Based Machine...

Area of Science:

• Artificial Intelligence
• Biomedical Informatics
• Cardiology

Background:

• Myocardial infarction (MI) diagnosis requires accurate and interpretable predictive models.
• Explainable Artificial Intelligence (XAI) techniques offer transparency in clinical decision-making.
• Biomarker identification is crucial for effective MI diagnosis and risk stratification.

Purpose of the Study:

• To develop an interpretable and accurate predictive model for MI using Explainable Boosting Machines (EBM).
• To identify and rank clinically relevant biomarkers for MI diagnosis.
• To enhance transparency in AI-driven diagnostic tools for clinical support.

Main Methods:

• Utilized a dataset of 1319 patient records from Iraq, including clinical and biochemical features.
• Applied preprocessing techniques such as one-hot encoding and normalization.
• Trained and tested an EBM model, evaluating performance with AUC, accuracy, sensitivity, and specificity.
• Assessed feature importance and employed partial dependence analyses for interpretability.

Main Results:

• The EBM model achieved high diagnostic performance: AUC 0.980, accuracy 96.6%, sensitivity 96.8%, specificity 96.2%.
• Troponin and CK-MB were identified as the most significant predictors of MI.
• Partial dependence plots revealed non-linear relationships between biomarkers and MI prediction.
• Local explanation plots confirmed the model's interpretable predictions for individual cases.

Conclusions:

• EBM offers a clinically valuable and ethical AI approach for MI diagnosis.
• The model's transparency supports biomarker prioritization and clinical risk stratification.
• Findings align with precision medicine and responsible AI principles, warranting further multi-center validation.