Intracardiac arrhythmia classification using neural network and time-frequency analysis

Time-frequency analysis (TFA) has been used to explore subtle differences in surface electrocardiographic signals. To determine whether TFA might improve the accuracy of automated intraventricular signal classification, the authors examined ten patients, each with ten cycles of sinus rhythm (SR) and ten cycles of ventricular tachycardia (VT). A time-frequency distribution (tfd) and a four line contour plot (CP) were generated for each cycle. A three layer backpropagation neural network (NN) was constructed in which three features were extracted from each CP cycle and used as inputs to the discriminant NN. One-half of the SR and VT cycles were utilized as a training set. Each patient-specific NN and a generalized NN bad an overall correct detection probability greater than 0.7. Using a NN as a classifier, time-frequency analysis improved specificity in arrhythmia classification without sacrificing sensitivity. It is concluded that combined neural network and TFA shows promise for automated intracardiac signal classification.<>