A STUDY OF MACHINE LEARNING APPLICATION IN BLOOD PRESSURE MEASUREMENT

Abstract

Effective continuous blood pressure monitoring is essential to preventing cardiovascular diseases. Pulse transit time (PTT) based blood pressure measurement method suffers from maintaining accuracy and re-calibration problem. The purpose of this study is to exam the performances of different machine learning methodologies, including linear regression, feed-forward neural network and recurrent neural network, in continuous blood pressure measurement. The experiment data was selected from MIMIC II online databases. Different initial settings of feed-forward neural network and recurrent neural network were tested. The experiment shows that the best error percentage is 7% using recurrent neural network method. Compared to the other machine learning methods, the proposed recurrent neural network structure with 5 hidden layers, 10 hidden neurons in each layer, has the best performance in tracing patient blood pressure among proposed machine learning structures.