Instantaneous Time-Frequency Features in Characterizing Ventricular Arrhythmias Using Empirical Mode Decomposition

摘要

Ventricular arrhythmias (VA) can lead to lethal conditions depending on their characteristics and temporal progression. Hence, it is essential to detect the type of VA and track its transitions over time to provide feedback in choosing appropriate therapy options. In this work, Empirical Mode Decomposition was used to extract intrinsic mode functions (IMFs) and construct the Hilbert energy spectrum (HS) from the 60-s long ECG segments during VAs. From the HS, instantaneous mean frequency and squared instantaneous bandwidth were extracted to track the progression of VAs. In addition, the energy ratio variance was computed from the IMFs. Using the extracted features, quantification of the performance was evaluated by a two-stage binary classification with a linear discriminant analysis based classifier and leave-one-out cross validation. A classification accuracy of 84% was achieved in classifying VT from VF, and 75% was achieved in classifying the correctly classified VF from the first stage into organized and disorganized VF.