BP-Net: Monitoring "Changes" in Blood Pressure Using PPG With Self-Contrastive Masking.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Physiological Signal Processing

Background:

• Deep learning models for blood pressure (BP) estimation from photoplethysmogram (PPG) signals face generalizability challenges.
• Continuous monitoring of acute BP changes is crucial for clinical applications like hypertensive emergencies.

Purpose of the Study:

• To develop a novel deep learning approach for detecting significant changes in systolic blood pressure (SBP) over time using PPG signals.
• To address the generalizability limitations of existing PPG-based BP estimation models.

Main Methods:

• Proposed a self-contrastive masking (SCM) model for pair-wise temporal comparisons within PPG signals.
• Introduced BP-Net, a model trained to detect SBP elevations or drops exceeding a threshold using PPG.
• Evaluated BP-Net on unseen subjects and across different datasets, comparing it with existing PPG-based BP estimation methods.

Main Results:

• BP-Net achieved 75.97% accuracy in detecting over-threshold SBP changes in unseen subjects.
• The model demonstrated improved generalizability across different datasets.
• BP-Net outperformed existing PPG-based models in detecting significant SBP fluctuations.

Conclusions:

• Framing BP estimation as detecting "changes" rather than absolute values enhances model generalizability.
• BP-Net offers a promising new approach for clinical BP monitoring using PPG signals.
• This work advances the use of PPG for reliable and generalizable BP change detection.