Continuous blood pressure prediction using pulse features and Elman neural networks

The present study designs an algorithm to improve the accuracy of continuous blood pressure (BP) prediction. Pulse wave transmission time has been widely used for continuous BP prediction. However, because of the limitation of the linear model and the complexity of signal acquisition traditional method is often troubled with low BP prediction accuracy. In this paper, Elman neural networks (Elmans) are used to construct the continuous blood pressure measurement model. Continuous measurement of blood pressure is achieved by a single channel pulse signal (PPG). Based on the time-related features of Elman, we apply a method using the feature input matrix by the pulse wave feature points at time t-1 and time t. We select the SBP and DBP at time t as the output matrix of the model; finally establish the Elman continuous blood pressure measurement model. Twenty physiological data segments of two hours selected from the MIMIC II database are used to evaluate the performance. Compared with straightforward use of the PWTT-based linear regression model and the back propagation neural network model, the proposed model achieves higher measurement accuracy.