Application of variational mode decomposition and ABC optimized DAG-SVM in arrhythmia analysis

Abstract

Automatic analysis of long-term electrocardiogram (ECG) recordings is crucial for timely and accurate diagnosis of life-threatening cardiovascular diseases. This article presents an efficient ECG classification scheme using variational mode decomposition approach. The method decomposes a time-domain input signal into various variational mode functions (VMFs). The VMD method adaptively decomposes an input signal into a number of modes to estimate their center frequencies, so that the band-limited modes can regenerate the input signal exactly. In this study, only mode-2 (M2) is used as morphological features and represented in reduced dimensions by employing principal component analysis (PCA). Further, the dynamic features (RR-intervals) are concatenated to constitute a feature set representing each heartbeat. The PCA method is employed to balance the impact of both the features exhibiting two different characteristics of an heartbeat i.e within the event and among the events. These extracted features of each heartbeat are further utilized for recognition into one of 16 heartbeat classes using artificial bee colony (ABC) optimized directed acyclic graph support vector machines (DAG-SVM). The proposed method is evaluated on the benchmark MIT-BIH arrhythmia database yielding an improved accuracy, sensitivity, positive predictivity and F-score of 98.72%, 98.72% and 98.72% respectively over the methodologies available in literature to the state-of-art diagnosis.