A Hybrid GCN-LSTM Model for Ventricular Arrhythmia Classification Based on ECG Pattern Similarity.

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Accurate differentiation between Ventricular Tachycardia (VT) and Ventricular Fibrillation (VF) is critical for patient outcomes.
• Traditional electrocardiogram (ECG) analysis relies on manual feature extraction, which can be time-consuming and prone to error.
• Deep learning models offer automated arrhythmia recognition, surpassing conventional methods.

Purpose of the Study:

• To develop an advanced deep learning model for automated classification of cardiac arrhythmias.
• To improve the accuracy in distinguishing VT and VF from non-ventricular rhythms.
• To leverage graph-based learning for capturing temporal dependencies in ECG data.

Main Methods:

• A novel model merging Graph Convolutional Networks (GCN) with Long Short-Term Memory (LSTM) networks was developed.
• A trainable weighted ϵ-neighborhood graph was employed to model similarities within ECG time series segments.
• The model was trained and evaluated for its classification performance on VT, VF, and non-ventricular rhythms.

Main Results:

• The GCN-LSTM model demonstrated substantial improvements in classifying VT, VF, and non-ventricular rhythms.
• This approach effectively captures complex patterns and similarities within ECG segments.
• The model surpasses traditional ECG analysis methods in accuracy and automation.

Conclusions:

• The developed GCN-LSTM model offers a highly accurate and automated solution for cardiac arrhythmia detection.
• This deep learning approach enhances the ability to differentiate critical arrhythmias like VT and VF.
• The findings highlight the potential of graph-based deep learning in cardiovascular signal analysis.