Confidence Interval and Respiratory Rate Estimations Using Smart Feature Fusion Based on Exact Gaussian Process

Respiratory rate (RR) monitoring is vital for people with respiratory conditions, especially the elderly. However, traditional respiration monitors only provide single-point estimates without confidence intervals (CIs). Therefore, the statistical variability of the estimate cannot be distinguished from the inherent variability caused by physiological processes. In this study, we present a novel method using exact Gaussian process regression (EGPR)-based multiple mix feature extraction (MMFE) and smart feature fusion using photoplethysmography (PPG) and electrocardiogram (ECG) signals to improve the reliability of RR and CIs estimation. To estimate CIs for accurate RR, we obtain the input data dimensions using power spectral (PS) features. After that, the multiphase feature extraction model is used to compensate for insufficient input data. Finally, we combine the four feature sets and selected features with high weights using adaptive neighbor component analysis (ANCA). The proposed EGPR methodology is currently the only method that can provide highly stable standard deviation and CIs. Therefore, the proposed model is a robust algorithm for rapidly estimating respiration rates in hospitals and medical centers, featuring significant estimation performance and low estimation mean absolute error (MAE), standard deviation (SD), and CIs. The proposed EGPR-based multiple multilevel features (MF) 0.98 breath per minute (bpm) based on the PPG (PMF-EGPR) model shows the lowest MAE compared with the other methods. Moreover, the proposed PMF-EGPR model represents 4.85 CI (bpm) for RR estimation.