ESTIMATION OF SYSTOLIC AND DIASTOLIC BLOOD PRESSURE USING PHOTOPLETHYSMOGRAPHY AND MACHINE LEARNING

Abstract

Suboptimal blood pressure is one of the biggest factors for health ailments around the world. Existing techniques for continuous monitoring of blood pressure include cuff devices which are difficult to use for continuous systolic and diastolic blood pressure measurements and may require medical supervision to gather accurate readings. In this study, we introduce a method for estimation of systolic blood pressure and diastolic blood pressure values using machine learning on the photoplythysmograph (PPG) signal collected from a sensor attached to the wrist. Subject specific influence on the PPG signal is determined and removed using signal filtering and proper normalization. Most important features extracted from PPG signal were phases and amplitudes of its Fourier transform. This data was given as input to artificial neural networks and other machine learning classifiers to predict the values of systolic and diastolic blood pressure. We collected multiple instances of data from 14 volunteers on various points of time in a day. Prediction obtained from the machine learning classifiers were in the form of regression and classification. All estimations show a good correlation between PPG signal and blood pressure. We were able to achieve mean error of 12% while predicting blood pressure using a trained model. This method involves the use of non-invasive wrist sensor and can potentially be used for wrist-based continuous blood pressure monitoring system. The system provides evidence for further investigation for continuous non-invasive blood pressure measurements using optical sensors.