Towards Accurate Estimation Of Cuffless And Continuous Blood Pressure Using Multi-Order Derivative And Multivariate Photoplethysmogram Features

Objective: Noninvasive estimation of cuffless and continuous blood pressure (CC-BP) is important for prevention and diagnosis of cardiovascular diseases. Many efforts have been made to estimate CC-BP, but current algorithms still dissatisfy the practical applications due to their limited accuracy and usability. While most previous studies used the features of hybrid pulse arrival time (PAT) and photoplethysmogram (PPG) for CC-BP estimation, this study investigated whether only using PPG features can estimate CC-BP efficiently and accurately.Methods: The PPG signals from 109 patients of the intensive care units were used to extract 65 features for CC-BP estimation. For comparison purpose, two previously reported hybrid feature sets with PAT and PPG indicators were also extracted from the PPG and electrocardiogram. A commonly used multiple linear regression algorithm was adopted for CC-BP estimation. To increase the usability of CC-BP estimation, a feature selection method was developed to choose the most critical and representative subset from the 65 PPG features based on their importance and stability in CC-BP estimation.Results: Our results demonstrated that the accuracy of the CC-BP estimation from 65 PPG features was significantly high in comparison to that of the hybrid PAT and PPG indicators (P < 0.05). When using the subset of the selected critical 13 PPG features, a comparable estimation accuracy could also be achieved as the hybrid PAT and PPG feature sets.Conclusion: The CC-BP estimation algorithm only based on PPG features would provide a convenient way to estimate CC-BP with a comparable accuracy, but personalized calibration should be optimized.