Generalized Deep Neural Network Model for Cuffless Blood Pressure Estimation with Photoplethysmogram Signal Only

Abstract

Due to the growing public awareness of cardiovascular disease (CVD), blood pressure (BP) estimation models have been developed based on physiological parameters extracted from both electrocardiograms (ECGs) and photoplethysmograms (PPGs). Still, in order to enhance the usability as well as reduce the sensor cost, researchers endeavor to establish a generalized BP estimation model using only PPG signals. In this paper, we propose a deep neural network model capable of extracting 32 features exclusively from PPG signals for BP estimation. The effectiveness and accuracy of our proposed model was evaluated by the root mean square error (RMSE), mean absolute error (MAE), the Association for the Advancement of Medical Instrumentation (AAMI) standard and the British Hypertension Society (BHS) standard. Experimental results showed that the RMSEs in systolic blood pressure (SBP) and diastolic blood pressure (DBP) are 4.643 mmHg and 3.307 mmHg, respectively, across 9000 subjects, with 80.63% of absolute errors among estimated SBP records lower than 5 mmHg and 90.19% of absolute errors among estimated DBP records lower than 5 mmHg. We demonstrated that our proposed model has remarkably high accuracy on the largest BP database found in the literature, which shows its effectiveness compared to some prior works.

Continuous Blood Pressure Estimation Using Exclusively Photopletysmography by LSTM-Based Signal-to-Signal Translation

“…It is reported from both studies that using combination features from ECG and PPG signals results in a comparatively better performance. However, the reported results could not surpass the other studies [6,7,[14][15][16] that produce remarkable results using PPG signal only. Despite the simplicity of PPG waveform, numerous features can be extracted from the time and frequency domain of an appropriate PPG signal.…”

“…Despite that, the number of subjects is also small compared to ours. Our latest work [ 14 ] presents 65 features and select 59 features as the input of four layers deep neural network (DNN) model to achieve the smallest SBP prediction error with the largest number of subjects. Ibtehaz et al [ 17 ] have the most similar objective to our study, which is trying to translate PPG signals to ABP signals.…”

“…DBP can be calculated following this equation: The general PPG waveform is quite simple and straightforward, yet not always informative [5]. Considering that the PPG represents how much amount the blood volume changes, taking its first and second derivatives can significantly help in evaluating the informative features in PPG waveform, such as the velocity and the accelerations of blood volume changes [4,14]. We take the derivatives of PPG waveform and stack it along to be the input of our model.…”

“…Our prior work [ 14 ] selects 59 out of 65 total features extracted from PPG signal as the input of the deep neural network (DNN) model to predict SBP and DBP. The resulting errors are exceptionally low from both SBP and DBP prediction.…”

Beat-to-Beat Continuous Blood Pressure Estimation Using Bidirectional Long Short-Term Memory Network

“…It was difficult to perform a fair comparison with other studies because the datasets used in the studies may differ significantly, and the validation methods also vary. Kachuee et al [ 19 ], Slapničar et al [ 20 ], and Hsu et al [ 21 ] used an online database named “Medical Information Mart for Intensive Care unit (MIMIC)” [ 22 ]. This database contains a large number of clinical data, including those of ECG, breathing, PPG, and BP.…”