A Novel Respiratory Rate Estimation Algorithm from Photoplethysmogram Using Deep Learning Model

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Signal Processing

Background:

• Respiratory rate (RR) is a crucial vital sign for diagnosing conditions like pneumonia.
• Manual RR counting is often inaccurate and inconsistent.
• Existing automated RR estimation methods require large data windows or lack precision.

Purpose of the Study:

• To develop a novel, continuous RR estimation method using photoplethysmogram (PPG) signals.
• To achieve accurate RR estimation with reduced window sizes and processing demands.
• To compare the performance of classical and deep learning algorithms for RR estimation.

Main Methods:

• Utilized the BIDMC and CapnoBase datasets for algorithm evaluation.
• Employed the Respiratory Rate Estimation (RRest) toolbox.
• Developed a deep learning model with convolutional and long short-term memory layers.
• Evaluated classical signal processing techniques and deep learning models.

Main Results:

• Optimal classical methods achieved a Mean Absolute Error (MAE) of 1.9 breaths/min on the BIDMC dataset.
• The novel deep learning model, using a 7s window, achieved an MAE of 2 breaths/min.
• The proposed model outperformed other deep learning algorithms using larger window sizes (16s, 32s, 64s).

Conclusions:

• Continuous RR estimation from PPG signals is feasible with reduced window sizes.
• The developed deep learning model offers a promising, computationally efficient approach for RR monitoring.
• Further research in signal processing can potentially improve PPG-based RR estimation accuracy.