Prediction of spontaneous ventricular tachyarrhythmia by an artificial neural network using parameters gleaned from short-term heart rate variability

Reducing casualties due to sudden cardiac death and predicting ventricular tachyarrhythmia (VTA), ventricular tachycardia (VT) or ventricular fibrillation (VF), is a key issue in health maintenance. In this paper, we propose a classifier that can predict VTA events using artificial neural networks (ANNs) trained with parameters from heart rate variability (HRV) analysis. The Spontaneous Ventricular Tachyarrhythmia Database (Medtronic Version 1.0), comprising 106 pre-VT records, 26 pre-VF records, and 126 control data, was used. Each data set was subjected to preprocessing and parameter extraction. After correcting the ectopic beats, data in the 5 min window prior to the 10 s duration of each event was cropped for parameter extraction. Extraction of the time domain and non-linear parameters was performed subsequently. Two-thirds of the database of extracted parameters was used to train the ANNs, and the remainder was used to verify the performance. Three ANNs were developed to classify each of the VT, VF, and VT + VF signals, and the sensitivities of the ANNs were 82.9% (71.4% specificity), 88.9% (92.9% specificity), and 77.3% (73.8% specificity), respectively. The normalized areas (Azs) under the receiver operating characteristic (ROC) curve of each ANNs were 0.75, 0.93, and 0.76, respectively.