Blood Pressure Estimation From Time-Domain Features of Oscillometric Waveforms Using Long Short-Term Memory Recurrent Neural Networks

This article presents a novel method for estimating systolic blood pressure (SBP) and diastolic blood pressure (DBP) from time-domain features extracted from oscillometric waveforms (OWs) using a long short-term memory (LSTM) recurrent neural network (RNN) method. First, we extract seven time-domain features from each cycle of OW, including the cuff pressure, the cardiac period, the trough-to-peak amplitude of OW, the time between the trough and the peak of the OW, the slopes of the oscillometric waveform envelope (OWE), and the maximum upslope of individual OWs. Second, we locate each feature vector in an noninvasive blood pressure (NIBP) record in one of three different phases (classes), namely, presystolic (PS), between systolic and diastolic (BSD), and after diastolic (AD), and form a target sequence. Then, we propose an LSTM-RNN approach to effectively learn the complex nonlinear relationship between the feature vector sequences and the target sequence. The SBP and DBP points are then obtained by mapping the beats at which the network output sequence switches from PS phase to BSD phase and from BSD phase to AD phase, respectively, to the deflation curve. Adopting a tenfold cross-validation scheme and using a database of 350 NIBP recordings gave an average mean error of -1.2 ± 5.9 mmHg for SBP and 1.8 ± 8.8 mmHg for DBP relative to reference values derived from a visual method of determining SBP and DBP. The proposed RNN-based approach uses all time-domain features available from each NIBP recording and can outperform traditional methods in blood pressure (BP) estimation.