Time domain variability of high resolution beat-to-beat recordings classified by neural networks

This study assesses the beat-to-beat variation of the time-domain QRS signal in normals and in patients with malignant ventricular arrhythmia using a multilayer perceptron neural network. High-resolution beat-to beat ECGs were recorded in 51 normals and in 46 CHD patients with inducible sustained VT (sVT). Per individual one variability vector was calculated. Total time-domain QRS variability, calculated as the sum of the elements of the variability vector (vector of standard deviations), was significantly higher in patients with inducible VT than in normals. To establish a user independent classification a multilayer perceptron was utilized on the variability vector. Solely based on the variability information of the QRS, the neural net gained a classification accuracy comparable to the standard time-domain approach. We conclude that patients with sVT show a significantly higher beat-to-beat variation inside the time domain QRS, possibly reflecting microvariations of the myocardial excitation.