Evaluation of the Photoplethysmogram-Based Deep Learning Model for Continuous Respiratory Rate Estimation in Surgical...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Physiological Monitoring

Background:

• Respiratory rate (RR) is a critical prognostic indicator, but its measurement often requires specialized equipment or indirect estimation.
• Photoplethysmogram (PPG) signals, commonly used for circulation monitoring, contain indirect information about intrathoracic pressure changes relevant to RR.
• Accurate and continuous RR monitoring is essential, particularly in intensive care settings.

Purpose of the Study:

• To evaluate the efficacy of various deep learning models for continuous and accurate respiratory rate estimation from PPG signals.
• To compare the performance of a novel Dilated Residual Neural Network against existing models for RR estimation.
• To assess the feasibility of using PPG for non-invasive RR monitoring in a clinical context.

Main Methods:

• A dataset of PPG signals was collected from 100 adult patients in a surgical intensive care unit.
• Public datasets (BIDMC, CapnoBase) were also analyzed.
• Seven deep learning models, including a Dilated Residual Neural Network, were trained and validated using 5-fold cross-validation, with Mean Absolute Error (MAE) as the primary metric.

Main Results:

• The Dilated Residual Neural Network achieved the best performance, with an MAE of 1.2628 ± 0.2697 on the BIDMC dataset and 3.1268 ± 0.6363 on the collected dataset.
• The study demonstrated the potential of PPG-derived deep learning models for RR estimation.
• Performance varied across datasets, highlighting the influence of data characteristics.

Conclusions:

• Estimating respiratory rate from PPG signals using deep learning remains a challenging task with limitations.
• The developed Dilated ResNet model shows promise for RR estimation, outperforming other evaluated models on specific datasets.
• Further research with larger, diverse datasets is needed to improve the accuracy and reliability of PPG-based RR monitoring models.