Automated Myocardial Infarction Screening Using Morphology-Based Electrocardiogram Biomarkers

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Ischemic heart disease (IHD) is a major global cause of death, often progressing to myocardial infarction (MI).
• Automated electrocardiogram (ECG) analysis offers a reliable and efficient alternative to manual interpretation for MI detection.
• Traditional 12-lead ECG acquisition can be inconvenient, necessitating simpler methods.

Purpose of the Study:

• To develop and validate an automated approach for classifying recent MI, past MI, and normal sinus rhythm (NSR).
• To investigate the efficacy of using a minimal set of ECG leads (leads I and II) for MI detection.
• To evaluate the performance of a random forest (RF) classifier trained on derived augmented limb leads.

Main Methods:

• Feature extraction from derived augmented limb leads using leads I and II.
• Training a random forest (RF) classifier on these extracted features.
• Employing five-fold cross-validation for performance evaluation.

Main Results:

• The RF classifier achieved a training accuracy of 97.9% (±0.008%) and a testing accuracy of 98%.
• The classifier built using features from all limb leads demonstrated superior performance compared to combinations.
• High sensitivity was reported for identifying recent and past MI classes.

Conclusions:

• The proposed automated ECG analysis method accurately detects recent and past MI using minimal leads.
• The approach is suitable for preventative healthcare, clinical screening, and remote monitoring applications due to its low complexity and high accuracy.
• The method's reliance on leads I and II makes it compatible with mobile and wearable devices.