Hybrid CNN-LSTM Model for Evaluating Heart Rate Variability from Pulse-to-Pulse Intervals

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Heart rate variability (HRV) is crucial for assessing autonomic function and cardiovascular health.
• Traditional HRV monitoring relies on electrocardiography (ECG), limiting non-invasive applications.
• Photoplethysmography (PPG) offers a potential non-invasive source for HRV analysis.

Purpose of the Study:

• To develop and validate a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model.
• To estimate R-R intervals (RRI) from PPG signals for non-invasive HRV monitoring.
• To assess the model's accuracy under controlled breathing conditions.

Main Methods:

• Utilized a hybrid CNN-LSTM deep learning architecture.
• Trained and validated the model using PPG data from 19 subjects.
• Collected data under controlled breathing rates (6, 18, 30 breaths/min).

Main Results:

• Achieved high correlation between predicted and actual RRI values.
• Demonstrated a reduction in Mean Absolute Error (MAE) from 62.01 ms to 49.62 ms.
• Observed accurate HRV capture at lower breathing rates, with underestimation of frequency components (VLF, LF, HF) at higher rates (p < 0.01).

Conclusions:

• The deep learning model shows promise for real-time, non-invasive HRV assessment using PPG signals.
• This approach provides a viable alternative to ECG-based HRV monitoring.
• Enables continuous cardiovascular monitoring via wearable devices for remote patient care and early disease detection.