Continuous Cuff-Less Blood Pressure Estimation Based on Combined Information Using Deep Learning Approach

Pulse transit time (PTT) is a promising way for continuous and unobtrusive blood pressure (BP) measurement. Many investigators made great efforts on cuff-less BP estimation. However, estimation of BP in clinic with a reliable accuracy is still a great challenge. In this paper, we propose a novel continuous blood pressure estimation method based on combined information including waveform information, artificial features and personal features. A 5 and 8 hidden layer deep neural networks had been constructed to learn the efficient and indetectable features associated with BP from the treated electrocardiogram (ECG) and photoplethysmogram (PPG) waveforms. Moreover, no calibration procedure is required in this approach. In our experiments, a total of 41267 beats from 85 subjects were performed in the 10-fold cross validation test to examine the accuracy of the proposed method. Besides, a supplementary experiment on another batch of subjects was performed for the robust test. We found that combined information was superior to the single feature in BP estimation. In addition, model 1 shows better performance in the 10-fold cross validation test. Meanwhile, model 2 with less hidden layers presented better robustness than model 1. The mean absolute difference (MAD) of systolic BP and diastolic BP for model 2 were 3.63 and 2.45 mmHg, respectively. In the comparison experiment, model 2 showed superiority in accuracy compared to the state-of-the-art methods especially in diastolic BP. Although the model in this paper need to be further improved, the presented advantages endow the proposed methods a feasible and promising application in future continuous cuff-less BP estimation.