Deep Convolutional Generative Adversarial Network with LSTM for ECG Denoising.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Electrocardiograms (ECG) are crucial for diagnosing cardiovascular diseases.
• Noise significantly interferes with ECG accuracy, necessitating effective denoising techniques.
• Accurate ECG signal reconstruction is vital for reliable physiological information extraction.

Purpose of the Study:

• To develop an advanced ECG denoising method using an improved Generative Adversarial Network (GAN).
• To enhance the accuracy of ECG signal processing in the presence of various noise types.
• To evaluate the proposed method's performance against existing state-of-the-art techniques.

Main Methods:

• Proposed an LSTM-DCGAN method, integrating Long Short-Term Memory (LSTM) layers with a Deep Convolutional Generative Adversarial Network (DCGAN).
• Implemented a network architecture combining convolutional layers for feature extraction and LSTM layers for time-series dependence.
• Validated the algorithm on the MIT-BIH Arrhythmia Database with noise from the MIT-BIH Noise Stress Test Database.

Main Results:

• The LSTM-DCGAN method successfully removed both single and mixed noise from ECG signals.
• Achieved an average SNRimp of 19.254 dB, RMSE of 0.028, and PRD of 10.350 for mixed noise removal.
• Outperformed DCGAN and LSTM-GAN methods in terms of higher SNRimp and lower RMSE and PRD scores.

Conclusions:

• The proposed LSTM-DCGAN approach demonstrates significant advantages for ECG processing.
• This method is highly effective in complex scenarios with mixed noise.
• The technique offers a promising solution for accurate ECG analysis and cardiovascular disease diagnosis.