A CNN Model for cuffless Blood Pressure Estimation from Nonlinear Characteristics of PPG Signals

Continuous and non-invasive monitoring of blood pressure (BP) is of high importance in preventing cardiovascular diseases. Currently, blood pressure control is widely performed by non-invasive cuffless-based devices. Major studies focus on extracting temporal and frequency characteristics of electrocardiograph (ECG) and photoplethysmograph (PPG) signals in improving the accuracy of BP estimation using intelligent algorithms. In this study, a novel framework based on a lightweight deep convolution neural network (CNN) is proposed, in which the sequences of the PPG are applied to the input. Also, a set of nonlinear features called recurrence quantification analysis (RQA) have been used to improve the ability of the model in estimating the systolic blood pressure (SBP) and diastolic blood pressure (DBP). The proposed framework was evaluated on the data extracted from the benchmark MIMIC-II dataset. The impact of the RQA features in BP estimation was investigated based on Spearman's statistics. The results obtained from the statistical analysis indicated that all RQA features could significantly distinguish between the levels of the actual values of both SBP and DBP (
$\mathrm{p} < 0.001$
). Moreover, The results of the CNN model showed that the set of augmented inputs from RQAs and PPG sequences could provide high accuracy in BP estimation. In particular, the Pearson's correlation coefficient (
$R$
) for the proposed model was achieved to be 0.94552 and 0.93916 for SBP and DBP, respectively.