SynchroSqueezing transform based cardiac disease classification

Abstract

Early identification of cardiovascular disease helps in treatment of heart related disease. Due to non-stationary nature of ECG and abundant recording, analysis of long term ECG recording by manual method posing great challenge. In this paper, an automatic diagnosis system is developed for detection and classification of cardiovascular disease through invariant feature extraction method. Often, ECG recording is contaminated by high frequency powerline interference and low frequency baseline wandering. Therefore, at first the recording must undergo noise removal treatment. In this study, an Adaptive denoising procedure is proposed which combines the synchrosqueezing transform and feature of wiener filter to achieve noise free recording. Then QRS complex detection followed by beat segmentation algorithm is applied for QRS beat template creation. Extraction of invariant feature from QRS beat template is proposed and such features used as input to train multiclass support vector machine for disease classification. The present study suggests that synchrosqueezing transform based diagnostic system achieves high classification accuracy.