Comparison of Different Machine Learning Techniques for the Cuffless Estimation of Blood Pressure Using Ppg Signals

Abstract

Blood pressure (BP) is currently measured using sphygmomanometers, and it is a crucial biomarker of a person's heart health. Hence, regular monitoring of blood pressure is important for early diagnosis and treatment. On the other hand, conventional blood pressure measurement devices discomfort patients, since the blood flow is cut off with the pressure exerted by the cuff while measuring systolic blood pressure. Nowadays, researchers are using different signals such as Electrocardiogram (ECG) and Photoplethysmography (PPG) to extract useful information like pulse arrival time (PAT) and pulse transit time (PTT) in order to estimate blood pressure without using a cuff. Two signals can be used simultaneously, but this method requires two sensors, which makes it expensive and unpractical. To overcome this, only PPG-based cuffless and continuous monitoring of blood pressure has been proposed in several studies. In this paper, in order to estimate systolic and diastolic blood pressure values, three different machine learning algorithms, i.e. Linear Regression (LR), Support Vector Regression (SVR) and Artificial Neural Networks (ANNs), were implemented using PPG signals and some other features such as body mass index (BMI), age, height and weight obtained from the patient. A new, short-recorded photoplethysmogram dataset was used for this purpose, and the results are compared in terms of mean absolute error.