A non-invasive continuous cuffless blood pressure estimation using dynamic Recurrent Neural Networks

Highlights

• Dynamic neural networks are used to estimate cuffless blood pressure.

• Folded and branched Recurrent Neural Networks (LSTM-NN and NARX-NN) were compared.

• Facilitates the use of blood pressure measurement with wearable technologies.

• Blood pressure estimation is performed for sequential time series signals.

Abstract

Cardiovascular diseases (CVD) have become the most important health problem of our time. High blood pressure, which is cardiovascular disease, is a risk factor for death, stroke, and heart attack. Blood pressure measurement is commonly used to limit blood flow in the arm or wrist, with the cuff. Since blood pressure cannot be measured continuously in this method, the dynamics underlying blood pressure cannot be determined and are inefficient in capturing symptoms. This paper aims to perform blood pressure estimation using Photoplethysmography (PPG) and Electrocardiography (ECG) signals that do not obstruct the vascular access. These signals were filtered and segmented synchronously from the R interval of the ECG signal, and chaotic, time, and frequency domain features were subtracted, and estimation methods were applied. Different methods of machine learning in blood pressure estimation are compared. Dynamic learning methods such as Recurrent Neural Network (RNN), Nonlinear Autoregressive Network with Exogenous Inputs Neural Networks NARX-NN and Long-Short Term Memory Neural Network (LSTM-NN) used. Estimation results have been evaluated with performance criteria. Systolic Blood Pressure (SBP) error mean ± standard deviation = 0.0224 ± (2.211), Diastolic Blood Pressure (DBP) error mean ± standard deviation = 0.0417 ± (1.2193) values have been detected in NARX artificial neural network. The blood pressure estimation results are evaluated by the British Hypertension Society (BHS) and American National Standard for Medical Instrumentation ANSI/AAMI SP10: 2002. Finding the most accurate and easy method in blood pressure measurement will contribute to minimizing the errors.

Introduction

Cardiovascular diseases (CVD) are among the leading diseases that cause human death today when the world population reaches 7.7 billion [1]. CVD’s primary risk factor is high blood pressure (BP), known as the silent killer. More than 10 million people are at high risk of BP [2], [3], [4]. This disease, which is mostly seen in the elderly population, has started to be seen in the lower age groups. Low physical activity [5], disorders in eating habits [6], increased consumption of animal fats [7], [8], [9] reduced the incidence of cardiovascular diseases to young and even children. BP, known as the pressure of blood on the vascular walls, can be neglected despite its easy measurement. Patients who are diagnosed with high BP try to keep BP within certain limits with the help of medication. If high BP diagnosis and treatment is not made in the early stages, it can have serious consequences such as heart attack, stroke, organ failure, and stroke [10], [11], [12]. Since BP is controlled by the sympathetic and parasympathetic nervous system, long-term blood pressure measurement is required in the diagnosis of the disease. BP devices used today are aneroid BP instruments that still use Korotroff sounds. Appropriate device research is ongoing for long-term BP measurements.

In the home and office environment, BP measurement vascular occlusion (VO) [13], [14] and non-vascular occlusion (VNO) systems are used for BP measurement. In the VO blood pressure measurement systems, the sleeve that connects to the wrist or arm is filled with air and compresses the vessels and nerves. Therefore, they are not suitable for long-term blood pressure measurements. VNO measurement systems, on the other hand, are more suitable for long-term blood pressure measurements since they do not apply any pressure. Studies in the literature also tend to vascular access VNO blood pressure systems. Physiological signals such as Electrocardiogram (ECG), Photoplethysmography (PPG) and Balistocardiography (BCG) are used in systems that do not block the vascular access. The Moens Korteweg [15], [16] equation shows that there is a relation between the movement of blood in the vessels and blood pressure. The rate of progression of the blood pressure pulse varies depending on the blood pressure. BP pulse progression time Pulse Transit Time (PTT) [17], [18], [19], [20], [21], [78], [79], [80], BP pulse vein velocity is called Pulse Wave Velocity (PWV) [22], [23]. These parameters are used in blood pressure measurements. Equation (1) shows Moens Korteweg.

$$\mathit{PWV}=\sqrt{\frac{h{E}_{\mathit{inc}}}{2\sigma R}}$$

R: Arterial radius, h: arterial wall thickness, σ: circumferential wall tension and Einc Yaong coefficient. According to the Moens-Korteweg equation, in the process of spreading blood pressure in the artery, it provides a perspective that predicts that the blood will travel faster in the artery in a hard pressure pulse. Arterial PWV is based on simple anatomical and physiological approaches. The arterial pulse wave velocity is obtained by dividing the length of the artery by the pulse of this length of the artery through the passage time (PTT). PTT starts with the opening of the aortic valve; it covers the time that ends when the pulse reaches the last measurement note. Fig. 1 shows the PTT measurement. It can be calculated with PWV Equation (2).

$$\mathit{PWV}=\frac{D}{\mathit{PTT}}$$

Variables such as vessel diameter, vessel thickness, blood density, and elasticity of vessels affect PTT. Since stress, age, and physiological factors change PTT, errors in blood pressure measurement may occur. VNO blood pressure measurement studies try to minimize errors that occur. However, all calculations and studies could not reach the clinical standard. Table 1 shows the comparison of occlusive and non-occlusive blood pressure measurement systems for the last five years.

The change in PTT from non-linear factors led researchers to different measurement methods. The PPG signal and BP signal are similar. PPG and ECG signals contain information about BP. BP can be measured by extracting features from ECG and PPG signals [48]. In the blood pressure measurement studies, time domain, morphological, and frequency domain features are extracted. The extracted features are used in machine learning methods [49] in training, and blood pressure is measured.

This paper aims to establish a continuous cuffless blood pressure measurement system by using a new method to extract features from ECG and PPG signals. In summary, ECG and PPG signals have been pre-processed. In this stage, both signals have been segmented, and then cleared of noises by moving averages method put into the moving averages (5, 10, 15 beats). Chaotic, time domain, and frequency domain features have been extracted from the signals cleared of noise and artifacts. For ECG and PPG signals, separate moving averages have been obtained, and the features have been extracted. Blood pressure has been measured by chaotic, time domain, and frequency domain features using SVR, NARX-NN, Coarse Tree, and Linear Regression methods. NARX-NN and LSTM-NN have dynamic artificial neural networks, RNN type, are used as an alternative to machine learning methods. Measurement results have been evaluated using Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE) performance criteria.