Cuff-less and Calibration-free Blood Pressure Estimation Using Convolutional Autoencoder with Unsupervised Feature...

Area of Science:

• Biomedical Engineering
• Cardiovascular Health
• Machine Learning in Healthcare

Background:

• Unobtrusive, continuous blood pressure (BP) monitoring is crucial for cardiovascular disease prevention.
• Existing cuff-less BP estimation methods often require calibration and manual feature engineering, limiting their practical application.
• Addressing these limitations is key to developing effective wearable BP monitoring devices.

Purpose of the Study:

• To validate the feasibility of using a convolutional autoencoder (CAE) for calibration-free, continuous BP estimation.
• To assess the performance of a CAE-based approach in unsupervised feature learning for BP prediction.
• To determine if the proposed method meets established clinical accuracy standards for BP monitoring.

Main Methods:

• Employed a convolutional autoencoder (CAE) for unsupervised feature extraction from physiological signals.
• Trained a regressor model utilizing CAE-derived features for continuous BP estimation.
• Utilized 10-fold cross-validation to rigorously evaluate model performance on data from 62 subjects.

Main Results:

• The developed CAE-based method demonstrated successful calibration-free BP estimation.
• Unsupervised feature learning effectively replaced manual feature selection.
• The accuracy of the predicted BP values met the Grade B standard set by the British Hypertension Society.

Conclusions:

• The proposed convolutional autoencoder method offers a viable solution for calibration-free, continuous blood pressure monitoring.
• Its ability to learn features unsupervisedly from signals enhances its applicability in wearable devices.
• This approach holds significant potential for improving cardiovascular disease management through accessible BP tracking.