A Robustness Evaluation of Machine Learning Algorithms for ECG Myocardial Infarction Detection.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Automatic electrocardiogram (ECG) analysis for myocardial infarction (MI) detection requires robust systems for clinical practice.
• Existing systems must balance reliability, complexity, and decision-making performance in realistic environments.

Purpose of the Study:

• To investigate the robustness and durability of an automatic ECG MI detection system without feature extraction.
• To evaluate the performance of supervised machine learning models under various noise conditions.

Main Methods:

• Utilized Support Vector Machine (SVM), k-nearest neighbors (KNN), and Random Forest (RF) models.
• Classified normal (NOR) and myocardial infarction (MI) using PTB database ECG records and MIT-BIH Noise Stress Test Database (NSTDB) noise.
• Employed inter-patient data separation and data normalization/segmentation.

Main Results:

• All tested models achieved performance metrics above 0.50 even at lower signal-to-noise ratio (SNR) levels.
• Models demonstrated acceptable performance across different noise types, indicating robustness.
• The investigated methods are considered sustainable and robust for specific noise forms.

Conclusions:

• The evaluated machine learning models are suitable for real-world ECG MI detection tools.
• These models show promise for use in challenging clinical circumstances with high noise levels.