Related Experiment Videos
ECG Signal Compression and Reconstruction Based on CNN-LSTM-Attention Model
Wenyan Liu1, Dongzhi Chen1, Ze Zhang2
1School of Information and Communication Engineering, North University of China, Taiyuan 030051, China.
Sensors (Basel, Switzerland)
|July 15, 2026
Summary
This study introduces a novel CNN-LSTM-Attention model for efficient electrocardiogram (ECG) data compression and reconstruction. The hybrid approach enhances accuracy in remote monitoring, outperforming existing methods.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Healthcare
- Signal Processing
Background:
- Cardiovascular diseases necessitate continuous monitoring via wearable electrocardiogram (ECG) devices.
- Large volumes of ECG data present challenges in transmission, storage, and real-time processing.
- Existing ECG compression methods, like compressed sensing and single deep learning models, have limitations in efficiency and feature extraction.
Purpose of the Study:
- To develop an efficient and accurate ECG compression and reconstruction model for remote monitoring applications.
- To address the limitations of traditional methods by integrating local waveform feature extraction and temporal dependency modeling.
- To improve the focus on critical ECG components like the QRS complex and ST segment for enhanced diagnostic accuracy.
Main Methods:
- A hybrid deep learning model combining Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and an Attention mechanism was developed.
- CNN layers were used to extract local ECG waveform features.
- LSTM networks were employed to capture long-term temporal dependencies in the ECG signals.
- An attention mechanism was integrated to prioritize and fuse salient diagnostic features.
Main Results:
- The proposed CNN-LSTM-Attention model demonstrated favorable comprehensive performance across compression ratios from 0.1 to 0.9 on the MIT-BIH Arrhythmia dataset.
- Performance metrics included a stabilized Percent Root-Mean-Square Deviation (PRD) of 10-12%, Signal-to-Noise Ratio (SNR) above 20 dB, and Root-Mean-Square Error (RMSE) below 0.25 mV.
- The hybrid model significantly outperformed single CNN and CNN-LSTM models in terms of reconstruction accuracy and stability.
Conclusions:
- The CNN-LSTM-Attention hybrid model offers a superior solution for ECG compression and reconstruction compared to existing methods.
- This approach effectively balances the need for data compression with the requirement for high-fidelity signal reconstruction in remote ECG monitoring.
- The model's ability to focus on key diagnostic features holds promise for improving the reliability of remote cardiovascular disease monitoring.
