Prediction of successful cardioversion in atrial fibrilation using wavelet analysis parameters and sample entropy

In a substantial number of patients atrial fibrillation (AF) recurs after successful electrical cardioversion, but at present there are no reliable clinical markers for confidently identifying the patients in which recurrence will occur within a short period of time. This study evaluates the predictive classification performance of some Discrete Wavelet Transform (DWT) indices in distinguishing recurrent and non-recurrent AF episodes. A validated database of 33 ECG recordings acquired from AF subjects undergoing cardioversion was used throughout the study, together with their known recurrence status at one month. The DWT was applied to these ECG recordings. Several parameters were extracted from the decomposition bands as potential features for predicting the recurrence of AF episodes. The estimated classification rate of the extracted features was evaluated using linear discriminant analysis (LDA). For a separate 11 registers training set and 22 registers testing set, the performance of the classifier testing set gave an estimated accuracy of 82%. We conclude that features extracted from sub-band decomposition of the ECG can provide some indicator of the likelihood of AF recurrence.