Improving EASI Model via Machine Learning Techniques and Regression Techniques

Abstract

We propose an approach to the interpretation of natural 12-lead Electrocardiography (ECG) is the standard tool for heart disease diagnose but measuring all 12 leads is often awkward and restricted by patient movement. In 1988, Gordon Dower has introduced the EASI-lead monitoring system that can reduce the number of electrodes from 10 downto 5 and also increases mobility of patients. In order to gain all 12-lead ECG back from the EASI-lead system, Dower’s equation was proposed then. Ever since various attempts have been explored to improve the synthesis accuracy. To find the best transfer function for synthesizing the 12-lead ECG from EASI-lead system, this paper presents a number of Machine Learning techniques including Support Vector Regression (SVR) and Artificial Neural Network (ANN). The experiments were conducted to compare the results from those Machine Learning methods to those of Linear Regression, Polynomial Regression, and Dower’s methods. The results have shown that the best performance amongst those methods with the least Root Mean Square Error (RMSE) values were obtained by SVR using spherical kernel function followed ANN, 3rd-order Polynomial Regression, Linear Regression and Dower’s equation, respectively.