UTransBPNet for cuffless and calibration-free blood pressure estimation under dynamic conditions

Area of Science:

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

Background:

• Accurate cuffless blood pressure (BP) estimation is challenging, especially during dynamic physiological changes.
• Existing methods often require calibration and struggle with intra-individual BP variability.
• There is a need for robust, calibration-free models for continuous BP monitoring.

Purpose of the Study:

• To introduce UTransBPNet, a novel, calibration-free deep learning model for cuffless BP estimation.
• To evaluate UTransBPNet's performance in tracking BP variations across diverse dynamic datasets.
• To analyze the impact of dataset characteristics on model generalizability.

Main Methods:

• Developed UTransBPNet, integrating a squeeze-and-excitation-enhanced Unet for short-range features and a transformer with cross-attention for long-range dependencies.
• Utilized high-resolution, multi-channel physiological signals for BP estimation.
• Conducted validation across dynamic datasets (Dataset_Drink, Dataset_Exercise, Dataset_MIMIC) in scenario-specific and cross-scenario settings.

Main Results:

• UTransBPNet demonstrated superior performance in tracking dynamic BP variations compared to existing models.
• Achieved high individual Pearson's correlation coefficients (e.g., 0.82±0.11 for SBP in Dataset_Exercise) and low mean absolute differences (e.g., 4.38 mmHg for SBP in Dataset_MIMIC).
• Highlighted the influence of dataset characteristics (distribution shift, imbalance, individual variability) on model performance and generalizability.

Conclusions:

• UTransBPNet represents a significant advancement in cuffless BP estimation, offering improved accuracy and robustness under dynamic conditions.
• The findings underscore the importance of well-curated datasets for ensuring the generalizability of BP estimation models.
• This work paves the way for more reliable, non-invasive BP monitoring in real-world healthcare scenarios.