Integrating recurrent neural network (RNN) and Navier-Stokes equations for noncontact blood pressure assessment

Abstract

The study is to establish the non-contact blood pressure measurement model. We propose a novel hybrid blood pressure assessment model. This model employs digital signal processing (DSP) to process the Imaging Photoplethysmography (iPPG) signal, utilizing Support Vector Machine (SVM) classification to determine the optimal signal location through three parameters. It is then compared with a PPG device. Through a CNN-LSTM model, it aims to reconstruct the ideal iPPG signal, transforming signals from the dermal layer into radial artery signals. Based on the Beer-Lambert law, the natural logarithm of iPPG intensity is proportional to blood flow velocity. Thus, a regression model for mean arterial pressure is developed in this work using heart rate and the intensity of iPPG signals. In conclusion, statistical test results confirm the validity of this study, indicating significant potential for the future development of noncontact blood pressure monitoring.