Energy-Efficient PPG-Based Respiratory Rate Estimation Using Spiking Neural Networks.

Area of Science:

• Biomedical Signal Processing
• Artificial Intelligence in Healthcare
• Physiological Monitoring

Background:

• Respiratory rate (RR) is a critical vital sign linked to overall health.
• Photoplethysmogram (PPG) signals are commonly used for extracting respiratory information.
• Existing RR estimation methods using signal processing and deep learning have limitations.

Purpose of the Study:

• To propose an end-to-end respiratory rate estimation approach using Spiking Neural Networks (SNNs).
• To minimize information loss during data conversion for RR estimation.
• To evaluate the efficacy and energy efficiency of the proposed SNN model.

Main Methods:

• Utilized a third-generation artificial neural network model: Spiking Neural Network (SNN).
• Employed PPG signal segments as direct inputs, converting them into sequential spike events.
• Incorporated feedback-based integrate-and-fire neurons for temporal information transmission.
• Evaluated the model on the BIDMC respiratory dataset with varying window sizes (16, 32, 64 s).

Main Results:

• Achieved mean absolute errors of 1.37 ± 0.04, 1.23 ± 0.03, and 1.15 ± 0.07 for 16, 32, and 64 s window sizes, respectively.
• Demonstrated superior energy efficiency compared to other deep learning models.
• Successfully estimated RR directly from PPG signals with minimal information loss.

Conclusions:

• Spiking Neural Networks show significant potential for accurate and efficient respiratory rate monitoring.
• The proposed SNN approach offers a novel and effective method for RR estimation from PPG signals.
• This advancement contributes to the field of biomedical signal processing and remote patient monitoring.