Non-invasive modeling of heart rate and blood pressure from a photoplethysmography by using machine learning techniques

Abstract

Cardiovascular diseases(CVD) is one of the major causes of deaths in the world, which may damage the endothelium cells which may lead to atherosclerosis and cardiac arrhythmias. Blood pressure is an important parameter and indicator in cardiovascular disease, patients with CVD who have multiple risk factors such as hypertension, stress, and obesity have been increasing. Therefore, it is important in the field of cardiovascular disease prevention to predict those at risk of cardiovascular diseases in the general population. Electrocardiogram is not suitable for wearable devices and PPG is a non-invasive, inexpensive, and convenient diagnostic tool for monitoring of heart and blood pressure. Here, we present a PPG (photoplethysmography) based non-invasive detection of heart rate and blood pressure, containing 1260 segments from 140 subjects an age range of 20 – 50 years. Data acquisition was carried out using the standard operating procedures. The present study investigates the photoplethysmography signal filtering of various noise removal, extraction of PPG morphological features and its derivatives which contain a blood circulatory system information, and finally derived forty five diagnostic engineered features. A novel signal processing technique was applied to extract salient pulse wave for heart rate and blood pressure detection. The subset of optimal features was extracted using feature extraction methods in relation to the physiology of heart rate and blood pressure processes. Prediction of heart rate, systolic blood pressure and diastolic blood pressure with the root mean squared error (RMSE) of 4.3 beats per minute, 5.7 mmHg and 5.5 mmHg between Sphygmomanometer and PPG from 10-fold cross-validation method.