Heart rate estimation for U-Net and LSTM models combining multiple attention mechanisms

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Accurate heart rate (HR) monitoring is crucial for disease prevention and health management.
• Photoplethysmography (PPG) signals are often corrupted by noise, challenging precise HR estimation.
• Existing deep learning models may struggle with complex noise conditions in PPG data.

Purpose of the Study:

• To develop a novel deep learning framework, DRL-Unet, for accurate HR estimation from noisy PPG signals.
• To enhance the robustness and precision of HR monitoring using advanced AI techniques.
• To validate the performance of the proposed model against conventional methods.

Main Methods:

• Integration of Denoising Autoencoder (DAE), U-Net architecture, and Long Short-Term Memory (LSTM) networks.
• Incorporation of Multi-Head Attention mechanism and Residual Network (ResNet) modules for improved performance.
• Validation on a public dataset from the IEEE Signal Processing Cup.

Main Results:

• DRL-Unet achieved superior performance compared to conventional deep learning models.
• Key performance metrics include MAE of 1.69 bpm, MSE of 3.05 bpm², RMSE of 1.71 bpm, MAPE of 1.15%, and Bias of 0.05 bpm.
• The model demonstrated high effectiveness under complex noise conditions.

Conclusions:

• DRL-Unet offers a significant advancement in accurate and reliable HR estimation.
• The proposed framework has the potential to improve early cardiovascular disease diagnosis.
• This technology can enhance continuous health monitoring capabilities.