Reconstruction of central arterial pressure waveform based on CNN-BILSTM

Abstract

Central arterial pressure is an important physiological indicator of the human cardiovascular system. Its noninvasive, continuous and accurate reconstruction and monitoring are essential to the evaluation, prevention and treatment of cardiovascular system diseases. However, it is difficult to improve the accuracy of noninvasive central arterial pressure reconstruction by traditional methods, which limits its clinical application and promotion. In this study, a model for reconstructing central artery pressure from radial arterial pressure waveforms based on a convolutional neural network and bi-directional long short-term memory (CNN-BiLSTM) was proposed. Central aortic pressure waveforms and radial arterial pressure waveforms were measured invasively before and after medication in 62 patients to evaluate the CNN-BiLSTM model for reconstructing central artery pressure. The CNN-BiLSTM model was compared with two traditional methods (autoregressive exogenous (ARX) and N-point moving average method (NPMA)) and four deep learning models (LSTM, BiLSTM, LSTM-BiLSTM and CNN-LSTM) in the mean absolute error (MAE), root mean square error (RMSE) and Spearman correlation coefficient (SCC). Experimental results showed that the proposed model achieved the best results on waveform reconstruction (MAE: 2.18 ± 0.13 mmHg, RMSE: 2.95 ± 0.16 mmHg). At the same time, a good reconstruction effect was obtained in the central arterial systolic pressure (RMSE: 3.34 ± 0.91 mmHg) and diastolic blood pressure (RMSE: 2.41 ± 0.18 mmHg). Therefore, the reconstruction model based on CNN-BILSTM is a potential method for noninvasive continuous monitoring of central arterial pressure.