Statistical comparison of time-frequency and neural network versus correlation waveform analysis

A combined time-frequency and neural network (TF/NN) method was demonstrated to be effective in ventricular tachycardia (VT) detection in a previous study. The same patient set was submitted to analysis by correlation waveform analysis (CWA) to assess the comparative performance of the TF/NN versus CWA method. CWA results were examined by three statistical techniques to give a fair and unbiased assessment. The 1st test was the Z-test. The original correlation coefficients (cc) were transformed via Fisher z transformation and the Z-test was applied to the transformed data. In the 2nd method, 95% tolerance limits were calculated for the transformed sinus rhythm (SR) and VT cc. In the 3rd method, variable sensing thresholds (TRs) were applied to the original cc. Results showed that CWA performed well under the 1st and 2nd tests. Under different sensing thresholds, CWA has similar performance using patient-specific thresholds for VT detection. However, TF/NN is superior to CWA when global thresholds are utilized