Development of an Improved Stacked U-Net Model for Cuffless Blood Pressure Estimation Based on PPG Signals

Area of Science:

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

Background:

• Accurate blood pressure monitoring is crucial for cardiovascular health management.
• Current cuff-based methods are invasive and inconvenient for continuous monitoring.
• Developing non-invasive, cuffless blood pressure estimation techniques is a significant medical challenge.

Purpose of the Study:

• To present an enhanced stacked U-Net deep learning model for cuffless blood pressure estimation.
• To improve the accuracy of non-invasive blood pressure measurements using only photoplethysmogram (PPG) signals.
• To address challenges in systolic blood pressure estimation through advanced model architecture.

Main Methods:

• Utilized an enhanced stacked U-Net deep learning architecture.
• Incorporated velocity plethysmogram input for improved feature extraction.
• Employed additive spatial and channel attention mechanisms to refine U-Net performance.
• Mitigated decoder mismatches within the U-Net architecture.

Main Results:

• Achieved mean absolute errors of 3.921 mmHg (systolic) and 2.441 mmHg (diastolic).
• Satisfied Grade A criteria by the British Hypertension Society and met AAMI accuracy standards.
• Outperformed existing PPG-only spectro-temporal methods.
• Demonstrated performance comparable to combined PPG and electrocardiogram (ECG) deep learning models.

Conclusions:

• The enhanced stacked U-Net model offers a promising solution for cuffless blood pressure monitoring.
• The model demonstrates potential for practical, low-cost, and non-invasive continuous blood pressure assessment.
• This technology could significantly improve remote patient monitoring and hypertension management.