CnnBoost: a multilevel explainable stacked ensemble framework for effective detection of Myocardial Infarction from...

Area of Science:

• Cardiology
• Artificial Intelligence
• Machine Learning

Background:

• Electrocardiogram (ECG) is a vital non-invasive tool for diagnosing cardiovascular diseases.
• Interpreting ECGs requires specialized expertise, creating a need for automated diagnostic assistance.
• Automated detection of heart abnormalities can improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop an explainable machine-learning framework for classifying myocardial infarction (MI) and other heart abnormalities using ECG images.
• To integrate deep learning and ensemble methods for enhanced ECG analysis.
• To improve the interpretability of AI models in cardiovascular diagnostics.

Main Methods:

• Utilized a dataset of South Asian ECG images across four classes: Normal, Abnormal, MI, and Previous History of MI.
• Developed CNNBoost, a Multilevel Explainable Stacked Ensemble model combining CNN spatial features with time-series data for XGBoost.
• Employed SHapley Additive Explanations (SHAP) for identifying diagnostically significant ECG leads, validated by a cardiologist.

Main Results:

• Achieved 99% accuracy, 98.58% AUC, and 96.47% AUPRC for MI classification.
• The explainability component successfully identified critical ECG leads, enhancing model trustworthiness.
• Demonstrated effective classification of various heart abnormalities.

Conclusions:

• The CNNBoost framework integrates deep learning and ensemble learning to enhance ECG classification accuracy and clinical relevance.
• The model's ability to learn spatial and temporal features improves interpretability and decision support for clinicians.
• Cardiologist validation supports the framework's potential for real-world healthcare applications, reducing misdiagnoses and aiding clinical decision-making.