Improving SpO₂ estimation accuracy under varying respiratory intensities using whale optimization and Gaussian...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Continuous oxygen saturation (SpO2) monitoring is vital for respiratory disorder management.
• Photoplethysmography (PPG) signals are commonly used for SpO2 estimation.
• Respiratory intensity variations can affect SpO2 measurement accuracy.

Purpose of the Study:

• To investigate the impact of varying respiratory intensities on SpO2 estimation.
• To develop and optimize a predictive model for enhanced SpO2 accuracy.
• To compare hyperparameter optimization algorithms for Gaussian Process Regression (GPR).

Main Methods:

• Collected 12,250 PPG signal segments from a laboratory simulator across SpO2 levels (80-100%) and six respiratory intensities.
• Developed a GPR model with hyperparameters optimized by Bayesian Optimization, Genetic Algorithm, Whale Optimization Algorithm (WOA), Grey Wolf Optimizer, and Particle Swarm Optimization.
• Evaluated model performance using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), and validated with real-world data.

Main Results:

• The WOA-optimized GPR (WOA-GPR) model demonstrated superior performance under varying respiratory conditions.
• WOA-GPR achieved MAE between 0.89-1.41 and RMSE between 0.63-0.86.
• Comparative analysis confirmed WOA-GPR's effectiveness over Support Vector Regression and Random Forest Regression.

Conclusions:

• The WOA-GPR model offers a promising approach for accurate real-time SpO2 estimation.
• Optimized GPR models can effectively mitigate the impact of respiratory intensity on SpO2 measurements.
• This method enhances the reliability of SpO2 monitoring in practical health applications.