Classification of ECG Arrhythmia Types Using 2D Recurrence Polar Maps and Deep Learning Techniques

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiac arrhythmias are a major cause of cardiovascular disease (CVD) complications like stroke and heart failure.
• Accurate electrocardiogram (ECG) diagnosis is vital but expert interpretation varies.
• Existing 1D ML/DL models for ECG analysis struggle with signal complexity.

Purpose of the Study:

• To develop novel 2D representations for 12-lead ECG signals.
• To improve automated cardiac arrhythmia detection using machine learning.
• To address limitations of 1D ECG analysis in capturing spatial-temporal signal dynamics.

Main Methods:

• Proposed two novel 2D ECG representations: Recurrence Polar Plots (RPP) and Cross Recurrence Polar Plots (CRPP).
• Utilized these 2D maps to train a multichannel neural network model.
• Classified four major arrhythmia types: Atrial Fibrillation (AF), Sinus Tachycardia (ST), Sinus Bradycardia (SB), and Ventricular Tachycardia (VT).

Main Results:

• Achieved high classification accuracies: 84.0% for AF, 94.5% for ST, 91.5% for SB, and 93.5% for VT.
• Demonstrated that RPP and CRPP effectively capture cyclic and directional ECG signal patterns.
• Showcased the model's ability to leverage complementary information from 2D maps.

Conclusions:

• RPP and CRPP offer a powerful approach for representing complex ECG signals.
• The unified multichannel model effectively detects major cardiac arrhythmias.
• This 2D mapping strategy significantly enhances automated ECG analysis for CVDs.