Cardiac Arrhythmia classification based on 3D recurrence plot analysis and deep learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Current AI methods for cardiac arrhythmia (CA) classification often use 1D or 2D ECG analysis, limiting the capture of complex spatiotemporal correlations.
• These methods may oversimplify ECG data as linear, failing to represent the nonlinear and nonstationary nature of cardiac electrophysiology.

Purpose of the Study:

• To develop and validate a 3D recurrence plot (RP)-based deep learning algorithm for improved CA classification.
• To explore nonlinear recurrent features in ECG and Vectorcardiography (VCG) signals for enhanced diagnostic performance.

Main Methods:

• Generated 3D ECG/VCG images from standard 12-lead ECG and 3-lead VCG signals.
• Employed a deep learning algorithm utilizing 3D recurrence plots to analyze nonlinear signal characteristics.
• Trained, validated, and tested the model on the PTB-XL dataset.

Main Results:

• The 3D RP-based method achieved high average F1 scores: 0.9254 for 3D ECG and 0.9350 for 3D VCG.
• Outperformed existing 1D (0.843 F1 score) and 2D (0.9015 F1 score) ECG-based CA classification methods.
• Demonstrated superior performance and visual interpretability compared to prior approaches.

Conclusions:

• The proposed 3D recurrence plot deep learning method effectively captures nonlinear, spatiotemporal features for accurate cardiac arrhythmia classification.
• This advanced technique offers significant improvements over traditional 1D and 2D analysis, showing promise for clinical application.