BpBLS: A Knowledge-Embedded Bi-Incremental Broad Learning System for Wearable Cuffless Blood Pressure Estimation

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Cuffless blood pressure (BP) measurement is crucial for the growing elderly population.
• Existing data-driven methods for cuffless BP estimation are accurate but computationally intensive, requiring full retraining for updates.
• This limitation hinders their practical application on wearable devices.

Purpose of the Study:

• To develop a novel, efficient, and flexible system for cuffless BP estimation using wearable biosignals.
• To address the time-consuming retraining issue of traditional machine learning models.
• To enhance the accuracy of BP estimation through knowledge embedding.

Main Methods:

• Proposed a Knowledge-embedded Bi-incremental Broad Learning System (BpBLS) with a flat structure for flexible updates.
• Introduced a Pulse Pressure Regularization (PPR) method to embed BP knowledge and improve accuracy.
• Validated the BpBLS on two large-scale datasets: CAS-BP and Aurora-BP.

Main Results:

• BpBLS achieved superior estimation accuracy on both datasets, with errors comparable to state-of-the-art methods.
• Demonstrated significant computational efficiency, with training times under ten seconds.
• Achieved an order of magnitude improvement in computational efficiency compared to traditional methods.

Conclusions:

• The BpBLS framework provides a flexible, lightweight, and highly efficient solution for cuffless BP measurement.
• This approach overcomes the limitations of traditional models, enabling faster updates and deployment on wearable devices.
• The study offers a promising direction for real-time, non-invasive BP monitoring.