Semantic ECG hash similarity graph.

Area of Science:

• Biomedical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Graph-based methods are advancing ECG time series analysis.
• Existing methods often miss global semantic correlations and are sensitive to noise.
• There's a need for robust graph structures in ECG signal processing.

Purpose of the Study:

• To develop a novel graph generation learning framework for ECG signals.
• To enhance the retrieval efficiency of graph-based deep learning models for ECG recognition.
• To capture intricate associations within and between ECG signals using semantic hash coding.

Main Methods:

• Developed the Semantic Hash Similarity Graph (SHSG) framework.
• Utilized semantic hash coding for supervised and unseen ECG signals.
• Constructed a global hash dictionary and assembled graph topology using Hamming similarity.
• Employed iterative optimization in the orthogonal domain for hash representation maintenance.
• Applied a fast Graph Convolutional Network (GCN) for ECG recognition validation.

Main Results:

• The SHSG framework effectively captures global semantic correlations in ECG signals.
• The proposed method demonstrates enhanced retrieval efficiency for graph-based deep learning.
• Experimental results on multiple ECG datasets confirm the robustness and effectiveness of the SHSG approach.
• The generated graph topology improves ECG recognition accuracy.

Conclusions:

• The novel graph generation framework significantly improves ECG signal analysis.
• Semantic hash coding offers a powerful approach to capture complex ECG signal relationships.
• The SHSG method provides a robust and effective solution for ECG recognition tasks.