Predicting hypertension using PPG sensor data and demographic factors: A machine learning approach.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Research

Background:

• Hypertension is a major global health issue requiring continuous monitoring.
• Traditional blood pressure monitoring methods have limitations, driving the search for improved solutions.
• Photoplethysmography (PPG) offers a potential non-invasive method for cardiovascular assessment.

Purpose of the Study:

• To evaluate the feasibility of using PPG signals and deep learning for hypertension detection.
• To compare the performance of different deep learning models (CNN-Attention, CNN-GRU, LSTM) for this task.
• To assess the value of accessible cardiovascular monitoring in resource-limited environments.

Main Methods:

• Analysis of 657 PPG recordings from 218 subjects at Guilin People's Hospital.
• Application of deep learning models: Convolutional Neural Network-Attention (CNN-Attention), CNN-Gated Recurrent Unit (CNN-GRU), and Long Short-Term Memory (LSTM).
• Data preprocessing included z-score normalization and augmentation, with an 80:20 train-test split.

Main Results:

• The CNN-GRU model achieved the highest performance with 75% accuracy and an AUC-ROC of 0.658.
• CNN-GRU demonstrated perfect recall (1.00) for identifying hypertensive cases.
• CNN-Attention achieved 61% accuracy, while LSTM showed the poorest performance.

Conclusions:

• Deep learning models, particularly CNN-GRU, show promise for accurate hypertension detection using PPG signals.
• Accessible cardiovascular monitoring via PPG is feasible and valuable, especially in resource-limited settings.
• This approach can contribute to better hypertension management and cardiovascular health surveillance globally.