Robust algorithm for arrhythmia classification in ECG using extreme learning machine.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Current artificial intelligence (AI) methods for arrhythmia classification, particularly neural networks, face challenges with slow learning speeds and unstable performance due to local minima.
• Improving the practicality of AI-driven arrhythmia detection necessitates advancements in learning speed and accuracy.

Purpose of the Study:

• To develop a novel arrhythmia classification algorithm that achieves fast learning speeds and high accuracy.
• To address the limitations of existing neural network-based approaches in terms of practical application.

Main Methods:

• The proposed algorithm integrates Morphology Filtering, Principal Component Analysis, and Extreme Learning Machine (ELM) for arrhythmia classification.
• It is designed to classify six distinct beat types: normal, left bundle branch block, right bundle branch block, premature ventricular contraction, atrial premature beat, and paced beat.

Main Results:

• The algorithm achieved an average sensitivity of 98.00%, average specificity of 97.95%, and average accuracy of 98.72% on the MIT-BIH arrhythmia database.
• Compared to back propagation neural networks (BPNN), radial basis function networks (RBFN), and support vector machines (SVM), the ELM-based algorithm demonstrated significantly faster learning times (290x, 70x, and 3x faster, respectively).

Conclusions:

• The novel algorithm effectively classifies arrhythmias with high accuracy and short learning times.
• The robustness of the algorithm was confirmed through comprehensive evaluation on the entire MIT-BIH arrhythmia database.