Support vector regression-based synthesis of 12-lead ECG system from the standard 5 electrode system using lead V1

Abstract

The standard 12-lead electrocardiogram (ECG) is a fundamental but very efficient clinical method for heart disease diagnosis. Measuring all 12 leads is often cumbersome and impractical especially on a long term monitoring. There have been ways to reduce the number of electrodes in ECG system also from 10 down to 5 or 6 electrodes and various regression methods were applied to derive back those 12-lead ECG. This paper presents how Support Vector Regression (SVR) was used to find a set of transfer function for deriving the 12-lead ECG from the standard 5-electrode setting using lead V1 system. All dataset used in this work has been obtained from PhysioNet database consisting of 4,810 samples. Five-fold cross-validation was applied to find the best parameter of SVR. Two kernel functions, RBF and ERBF, have been explored and evaluated in this work. The experiments strongly presented that SVR methodology was worth considerate for synthesizing the 12-lead ECG signals from the standard 5-lead electrode system using V1. The results also showed ERBF kernel function gave better RMSE (Root Mean Square Error) than RBF kernel function in the case here.