Machine Learning-Based Respiration Rate and Blood Oxygen Saturation Estimation Using Photoplethysmogram Signals.

Area of Science:

• Biomedical Engineering
• Medical Devices
• Signal Processing

Background:

• Continuous monitoring of respiratory rate (RR) and oxygen saturation (SpO2) is vital for managing patients with cardiac, pulmonary, and surgical conditions.
• Photoplethysmogram (PPG) signals are increasingly recognized for their potential in evaluating RR and SpO2.
• Existing methods require further optimization for reliable and cost-effective patient monitoring.

Purpose of the Study:

• To develop and validate a novel machine learning approach for estimating RR and SpO2 from PPG signals.
• To identify optimal features from PPG signals for accurate vital sign estimation.
• To compare the performance of various machine learning models for RR and SpO2 prediction.

Main Methods:

• Extraction of meaningful features from PPG signals using established techniques.
• Application of a feature selection approach to reduce computational complexity and prevent overfitting.
• Training and evaluation of 19 distinct machine learning models for RR and SpO2 estimation.
• Selection of the best-performing regression model, specifically Gaussian process regression.

Main Results:

• The Gaussian process regression model demonstrated superior performance in estimating both RR and SpO2.
• Achieved a Mean Absolute Error (MAE) of 0.89 and Root-Mean-Squared Error (RMSE) of 1.41 for RR.
• Achieved an MAE of 0.57 and RMSE of 0.98 for SpO2.
• The proposed system represents a state-of-the-art method for reliable PPG-based vital sign estimation.

Conclusions:

• The developed machine learning system reliably estimates RR and SpO2 using PPG signals.
• This approach offers a potentially cheaper and less intrusive method for continuous patient monitoring.
• Successful derivation of RR and SpO2 from PPG could significantly enhance patient self-management and healthcare accessibility.