SIGNAL PROCESSING AND MACHINE LEARNING FOR BLOOD PRESSURE CLASSIFICATION USING ONLY THE ECG SIGNAL

Abstract

Continuous reading of vital signs in the Intensive Care Unit is a major role for the physician, which allows him to intervene in a timely manner. Thus, continuous blood pressure measurement remains a difficult task as long as it is based on using a mercury device or other wide varieties of methods. The approach of this research is based on classifying blood pressure records obtained from the analysis of the Electrocardiogram (ECG) solely using signal processing techniques. The analysis starts with Butterworth filtration of the ECG signal. Following that trend removal and normalization of the signal takes place before extracting 27 features. Feature selection methods are applied to reduce the number of features to the most dominant ones, and as a result the number of features was reduced to 10. The final results point to a high accuracy of 98.18% using a support vector machine (SVM) classifier. Other classifiers like artificial neural networks (ANN) and Bayesian naïve (BN) classifiers were also used but gave a less accuracy of 96.5% and 96.08%, respectively