[Automatic detection of ventricular and supraventricular wide QRS arrhythmias using complex of morphological criteria and algorithms].

Aim: The aim of study is a detection of ventricular and supraventricular wide QRS arrhythmias using complex of morphological criteria and algorithms by method of automatic analysis.

Materials And Methods: For 100 patients (m/f - 61/39, Me (min; max) - 44.5 (10; 85) years) of researched group the analysis of 14306 single wide ectopic complexes (QRS 120-230 ms) has been done. Wide complexes include 11028 (77%) ventricular complexes and 3278 (23%) supraventricular complexes represented by 145 different forms of QRS. For verification of arrhythmias origin transesophageal ECG recording and endocardial electrophysiological study were done. The control group included 59 patients (m/f - 25/34, Me (min; max) - 49.5 (14,85) years) with 720 wide QRS, including 467 (65%) ventricular and 253 (35%) supraventricular complexes represented by 86 forms of QRS. The criteria Drew B.J., Scheinman M.M. (1995); Wellens H.J. (1978), RWPT II (Pava LF, 2010) and the algorithms of Brugada P. (1991); Bayesian (2000); Vereckei A. (2008) were used to evaluate sensitivity, specificity and diagnostic accuracy of wide QRS complexes recognition one by one and together, using the method of Wald sequential automatic analysis (KT Result3, CJSC INCART, Russia) and method of artificial neural networks.

Results: The best results for the detection of ventricular arrhythmias algorithms were demonstrated by the  Brugada  P., Drew  B.J., Scheinman  M.M. algorithm (sensitivity 86.43%, specificity 66.73%, diagnostic accuracy 82.14% in the study group, sensitivity 81.80%, specificity 73.12%, diagnostic accuracy 78.75% in the control group), and the Bayesian algorithm (sensitivity 87.81%, specificity 73.62%, diagnostic accuracy 84.72% in the study group, sensitivity 83.30%, specificity 77.08%, diagnostic accuracy 81.11% in the control group). A complex analysis of the Wald method recognized ventricular arrhythmias in the research group with sensitivity 83.11%, specificity 83.65%, diagnostic accuracy 83.23% and in the control group with a sensitivity 83.51%, specificity of 84.58% and diagnostic accuracy 83.89%. Artificial neural networks recognized ventricular arrhythmias with sensitivity 91.43%, specificity 91.30% and diagnostic accuracy 91.39% in the control group and with sensitivity 97.06%, specificity 99.39% and diagnostic accuracy 97.6% in the research group.

Conclusion: Automatic analysis allows obtaining simultaneously the results of each algorithms/criteria and in combination. It significantly reduces the doctor's work in assessing of amplitude-time characteristics of the complexes. Using artificial neural networks increases the accuracy of of ventricular and supraventricular arrhythmias recognition.