Unraveling blood pressure estimation with a deep learning approach using multiple embeddings.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiovascular Monitoring

Background:

• Accurate blood pressure (BP) monitoring is crucial for cardiovascular health management.
• Existing BP measurement methods often require invasive procedures or frequent calibration.
• Continuous, non-invasive BP monitoring remains a significant clinical challenge.

Purpose of the Study:

• To develop and validate a calibration-free machine learning framework for estimating systolic BP (SBP) and diastolic BP (DBP) using pulse arrival time (PAT).
• To enhance the accuracy and reliability of BP estimation through advanced feature engineering and deep learning techniques.
• To assess the clinical applicability and generalizability of the proposed framework across diverse datasets.

Main Methods:

• Utilized pulse arrival time (PAT) derived from electrocardiogram (ECG) and photoplethysmography (PPG) signals.
• Incorporated similarity-based features using Euclidean and Manhattan distance matrices for enhanced pattern recognition.
• Employed an attention-guided convolutional neural network (CNN) for BP prediction.
• Validated the framework on three independent datasets: Cabrini Hospital, PTT PPG, and MIMIC-II.

Main Results:

• Achieved high R-squared values for SBP (0.89-0.95) and DBP (0.89-0.94) across datasets.
• Reported low mean absolute errors for SBP (1.31-6.45 mmHg) and DBP (0.98-2.92 mmHg).
• Met the Advancement of Medical Instrumentation (AMI) standard and achieved British Hypertension Society (BHS) Grade 'A'/'B' on multiple datasets.

Conclusions:

• The proposed calibration-free machine learning framework offers accurate and reliable BP estimation.
• The framework demonstrates strong generalizability, real-time compatibility, and clinical applicability.
• This approach provides a promising solution for continuous, comfortable, and non-invasive BP monitoring.