Detection of inter-patient left and right bundle branch block heartbeats in ECG using ensemble classifiers

Area of Science:

• Cardiology
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Left bundle branch block (LBBB) and right bundle branch block (RBBB) can mask electrocardiogram (ECG) findings and indicate underlying cardiac pathology.
• Accurate detection of LBBB and RBBB is crucial for managing cardiac diseases.
• Inter-patient heartbeat classification requires robust systems for predicting unknown data.

Purpose of the Study:

• To propose an inter-patient heartbeat classification method for accurate detection of LBBB and RBBB from normal beats (NORM).
• To evaluate the performance of a novel classification system in a clinical setting.

Main Methods:

• A combination of three classifiers: minimum distance, weighted linear discriminant, and linear support vector machine (SVM).
• Majority voting strategy to combine class labels from individual classifiers.
• Cross-validation for optimal parameter selection and assessment of lead configurations.

Main Results:

• A two-lead configuration demonstrated superior classification results compared to a single-lead configuration.
• Improved heartbeat classification performance was achieved by constructing classifiers for each heartbeat pair.
• Sensitivity and positive predictive values were reported for LBBB and RBBB detection.

Conclusions:

• A multi-classifier ensemble method using inter-patient data shows promising classification performance.
• The proposed approach has potential for clinical utility in differentiating LBBB and RBBB from NORM in new patients.