rU-Net, Multi-Scale Feature Fusion and Transfer Learning: Unlocking the Potential of Cuffless Blood Pressure...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiovascular Health

Background:

• Cuff-based blood pressure measurement is invasive and inconvenient.
• Developing accurate cuffless blood pressure monitoring is a significant challenge.
• Existing methods often lack accuracy and reliability for widespread clinical use.

Purpose of the Study:

• To introduce an innovative deep-learning model for cuffless blood pressure estimation.
• To achieve state-of-the-art performance using photoplethysmography (PPG) and electrocardiogram (ECG) signals.
• To enable accurate and reliable wearable blood pressure monitoring devices.

Main Methods:

• Developed a novel rU-Net deep-learning architecture, fusing U-Net and ResNet.
• Utilized short-time Fourier transform (STFT) for multi-scale feature extraction.
• Incorporated multi-head attention mechanisms and demographic parameters for enhanced accuracy.
• Employed a fine-tuning transfer learning approach for calibration-free data.

Main Results:

• Achieved state-of-the-art performance on the PulseDB dataset.
• Exceeded AAMI standards and earned a BHS Grade A rating on calibration-based data (SBP MAE ± std: 4.49 ± 4.86 mmHg, DBP MAE ± std: 2.69 ± 3.10 mmHg).
• Demonstrated exceptional accuracy on calibration-free data with 10% transfer learning (SBP MAE ± std: 4.14 ± 5.01 mmHg, DBP MAE ± std: 2.48 ± 2.93 mmHg).

Conclusions:

• The proposed deep-learning model offers a highly accurate and reliable solution for cuffless blood pressure estimation.
• The rU-Net architecture and transfer learning approach effectively address challenges in blood pressure monitoring.
• This study paves the way for advanced wearable devices for continuous, non-invasive blood pressure tracking.