Enhancing myocardial infarction detection with vectorcardiography: fusion-based comparative analysis of machine...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Myocardial infarction (MI) detection is crucial for timely treatment.
• Vectorcardiography (VCG) offers unique spatial insights into cardiac electrical activity, surpassing traditional electrocardiography.
• Limited studies compare machine learning (ML) techniques for MI detection using VCG data.

Purpose of the Study:

• To develop and evaluate a novel VCG processing methodology for automated MI detection.
• To compare the performance of various ML algorithms on VCG features.
• To enhance MI detection reliability through ensemble methods.

Main Methods:

• Extracted domain knowledge VCG features monitoring morphological changes.
• Applied 210 ML configurations including KNN, SVM, ANN, Random Forest, and ensemble methods.
• Utilized a stacking ensemble strategy for decision-level fusion of high-performing models.

Main Results:

• Stacking-based decision-level fusion achieved 95.55% accuracy.
• High sensitivity (97.70%) and specificity (86.25%) were reported.
• Positive predictive value reached 96.86% and F1-score was 97.27%.

Conclusions:

• Decision-level fusion via stacking significantly improves MI detection performance.
• The proposed method enhances the reliability of MI detection from VCG.
• This approach supports cardiologists in clinical decision-making for MI patients.