A cascaded classifier for multi-lead ECG based on feature fusion

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) is vital for diagnosing heart disorders.
• Current automatic diagnosis methods often use single or two-lead ECG, limiting feature extraction.
• Multi-lead ECG analysis offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a cascaded classification system for enhanced ECG beat classification using multi-lead signals.
• To improve the performance of automatic heart disorder diagnosis by leveraging comprehensive lead information.

Main Methods:

• A novel feature fusion method combining ten signal features from all 12 ECG leads.
• Implementation of a cascaded classifier using Random Forest (RF) and Multilayer Perceptron (MLP).
• Application of Principal Component Analysis (PCA) for feature space dimension reduction.

Main Results:

• The proposed system achieved high average accuracies: 99.3% (MIT-BIH Arrythmia), 99.8% (QT Database), 97.6% (MIT-BIH Supraventricular Arrhythmia), and 99.6% (INCART Database).
• The method demonstrated strong generalization capability across diverse datasets.
• Performance surpasses most existing state-of-the-art methods in ECG beat classification.

Conclusions:

• The cascaded classification system effectively utilizes multi-lead ECG data for accurate heart disorder diagnosis.
• The novel feature fusion and classification approach significantly enhances ECG beat classification performance.
• This method shows promise for robust and generalizable automatic diagnosis of cardiac conditions.