Comparison of four techniques for recognition of ventricular fibrillation from the surface ECG.

Four ventricular fibrillation (VF) detection techniques were assessed using recordings of VF to evaluate sensitivity and VF-like recordings to evaluate specificity. The recordings were obtained from Coronary Care Unit patients. The techniques were: threshold crossing intervals (TCI); peaks in the autocorrelation function (ACF); signal content outside the mean frequency (VF-filter); and signal spectrum shape (spectrum). Using 70 extracts, each 4 s long, from VF recordings, the VF filter achieved a sensitivity of 77 per cent; the ACF, TCI and spectrum algorithms had sensitivities of 67, 53 and 46 per cent, respectively. Susceptibility to false alarms was assessed using 40 extracts from VF-like recordings. The TCI algorithm was the most specific (93 per cent), while the spectrum, VF filter and ACF algorithms had specificities of 72, 55 and 38 per cent, respectively. The TCI algorithm achieved overall sensitivity of 93 per cent and specificity of 60 per cent. The spectrum, VF filter and ACF algorithms had overall sensitivities of 80, 93 and 87 per cent, and overall specificities of 60, 20 and 0 per cent, respectively.