PPG-Based Continuous BP Waveform Estimation Using Polarized Attention-Guided Conditional Adversarial Learning Model.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiovascular Physiology

Background:

• Blood pressure (BP) waveform analysis provides critical cardiovascular insights.
• Accurate BP monitoring is essential for diagnosing and managing cardiovascular diseases.
• Current methods for BP measurement can be invasive or intermittent.

Purpose of the Study:

• To develop a novel attention-guided conditional generative adversarial network (cGAN) for estimating BP waveforms from photoplethysmography (PPG) signals.
• To evaluate the performance of the proposed PPG2BP-cGAN model against state-of-the-art techniques.
• To assess the potential of the model for continuous, cuffless BP waveform measurement.

Main Methods:

• A conditional generative adversarial network (cGAN), named PPG2BP-cGAN, was designed with a UNet3+-based generator incorporating spatial-temporal attention and a PatchGAN discriminator.
• The generator utilized full-scale skip connections and a modified polarized self-attention module.
• The model was pre-trained on a dataset of 683 subjects and validated on an independent public dataset.

Main Results:

• The PPG2BP-cGAN model demonstrated superior BP waveform prediction performance compared to existing state-of-the-art methods.
• On the MIMIC dataset, the model achieved a root mean square error (RMSE) of 3.54 and a mean absolute error (MAE) of 2.86.
• Systolic and diastolic BP estimation errors met the AAMI standard (Systolic: 0.72 ± 4.34 mmHg; Diastolic: 0.41 ± 2.48 mmHg).

Conclusions:

• The proposed PPG2BP-cGAN method offers a highly accurate and reliable approach for estimating BP waveforms from PPG signals.
• The model's performance on independent datasets highlights its potential for non-invasive, continuous BP monitoring.
• This technology could significantly advance the field of cardiovascular health monitoring and management.