Conv-RGNN: An efficient Convolutional Residual Graph Neural Network for ECG classification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electrocardiogram (ECG) analysis is vital for diagnosing cardiovascular diseases (CVDs).
• Current deep learning methods often overlook spatial relationships in 12-lead ECGs, treating them as simple sequences.
• Representing ECGs in non-Euclidean space better captures intrinsic lead relationships.

Purpose of the Study:

• To develop an innovative deep learning method for ECG classification.
• To improve automated CVD diagnosis by incorporating both temporal and spatial ECG features.
• To enhance the diagnosis of CVDs using a novel graph neural network approach.

Main Methods:

• Proposed Convolutional Residual Graph Neural Network (Conv-RGNN) for ECG classification.
• Mapped 12-lead ECG signals to a graph structure to capture inter-lead spatial relationships.
• Utilized a convolutional neural network with attention for temporal feature extraction and a residual graph neural network for spatial feature extraction.

Main Results:

• Conv-RGNN demonstrated high competitiveness across multiple datasets (two multi-label, one single-label).
• The method achieved exceptional parameter efficiency, fast inference speed, and robust performance.
• Experimental results validate the effectiveness of integrating spatial and temporal information.

Conclusions:

• Conv-RGNN presents a promising and feasible approach for intelligent ECG diagnosis.
• The method is particularly suitable for resource-constrained environments.
• This work advances automated CVD diagnosis through a novel graph-based deep learning framework.