Interpretable Detection and Location of Myocardial Infarction Based on Ventricular Fusion Rule Features.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Myocardial infarction (MI) is a leading cause of death globally.
• Accurate and interpretable MI detection is crucial for timely clinical intervention.
• Electrocardiograms (ECG) are a primary diagnostic tool for MI, but interpretation can be complex.

Purpose of the Study:

• To develop an accurate and interpretable classification method for myocardial infarction detection.
• To integrate rule-based ECG features with ventricular activity features for improved diagnostic performance.
• To evaluate the efficacy of dimensionality reduction techniques on ventricular activity features.

Main Methods:

• Extracted rule features from ECG components (Q wave, ST segment, T wave) based on clinical MI criteria.
• Utilized all QT segment samples as ventricular activity features.
• Compared Discrete Wavelet Transform (DWT), Principal Component Analysis (PCA), and Locality Preserving Projections (LPP) for ventricular feature dimensionality reduction.
• Fused rule and processed ventricular features into a comprehensive feature vector.
• Employed eXtreme Gradient Boosting (XGBoost) for MI classification.

Main Results:

• Achieved an overall accuracy rate of 99.86% on the Physikalisch-Technische Bundesanstalt (PTB) database.
• Demonstrated the effectiveness of combining rule-based and ventricular activity features.
• Validated the performance of feature extraction and fusion techniques.

Conclusions:

• The proposed method offers a highly accurate and interpretable approach to myocardial infarction detection.
• The integration of clinical rule features and advanced signal processing enhances diagnostic capabilities.
• This approach holds significant potential for improving clinical decision-making in cardiology.