A new deep learning framework based on blood pressure range constraint for continuous cuffless BP estimation.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiovascular Health

Background:

• Traditional blood pressure (BP) measurement methods are invasive or cuff-based, limiting continuous monitoring.
• Accurate and continuous BP monitoring is crucial for early cardiovascular disease detection.

Purpose of the Study:

• To propose a novel cuffless BP estimation framework, RFPASN, integrated with BP range constraint.
• To enhance the accuracy and robustness of continuous BP monitoring using photoplethysmography (PPG) signals.

Main Methods:

• Developed a Receptive Field Parallel Attention Shrinkage Network (RFPASN) utilizing multi-scale convolutions for long-term PPG signal dynamics.
• Implemented a parallel mixed domain attention module and soft thresholding for feature screening and enhancement.
• Integrated a BP range constraint to improve the accuracy and stability of BP predictions.

Main Results:

• The RFPASN framework achieved Mean Absolute Errors (MAE) of 1.63/1.59 mmHg for diastolic blood pressure (DBP) and 2.26/2.15 mmHg for systolic blood pressure (SBP) on the MIMIC-II database.
• Demonstrated superior performance compared to existing state-of-the-art methods for BP estimation.
• Validated on a diverse dataset including normal, hypertensive, and hypotensive individuals.

Conclusions:

• The proposed RFPASN framework offers a promising, accurate, and robust solution for cuffless continuous blood pressure monitoring.
• The integration of attention mechanisms and BP range constraint significantly improves prediction accuracy.
• This technology has the potential to revolutionize cardiovascular health monitoring.