An Efficient Method for Computer-Aided Diagnosis of Cardiac Arrhythmias

Abstract

AbstractIn this chapter, an efficient features representation and machine learning methods are combined and developed to process the ECG signals. Initially, the raw heartbeats are pre-processed for eliminating various kinds of noises inherited within them. Consequently, the QRS-wave is located by applying Pan-Tompkins (PT) technique within the signals. Following the QRS-wave localization, a rectangular window of fixed size is selected for segmenting the heartbeats. Then, the empirical mode decomposition (EMD) algorithm is utilized for extracting the time domain information from heartbeats as features. Few coefficients are selected for an efficient representation of heartbeats using principal component analysis (PCA) which further reduces the complexity during processing using classifier. These output coefficients represent the characteristics of individual heartbeats and supports in distinguishing between them based on their morphology. Further, the R-peak to R-peak information between heartbeats are captured and concatenated with the output time-frequency coefficients. As a result, this final feature vector represents each heartbeat that are applied to support vector machine (SVM) model for recognizing these feature representations into corresponding classes of heartbeats. The classifier performance is also enhanced as its parameters are employed by employing the particle swarm optimization (PSO) algorithm under patient specific scheme. The proposed methodology is validated over Physionet database and the output of classifier model are compared to the labels of corresponding heartbeats of the database to formulate the results. The experiments conducted reported a higher overall accuracy of 95.86% over existing state-of-art methods.KeywordsElectrocardiogram (ECG)ArrhythmiasEmpirical mode decompositionR to R waveSupport vector machines