Determination of the etiology of wide-QRS tachycardias using an artificial neural network

The authors describe the development of a neural network designed to differentiate the etiology of wide-QRS tachycardias using a twelve-lead electrocardiogram (ECG). Four different etiologies of tachycardia were considered: coronary artery disease, right ventricular dysplasia, antidromic circus movement tachycardia, and idiopathic ventricular tachycardia. In 148 ECGs, 22 variables were collected. A large number of combinations were tested. In all cases at least 52% and up to 60% of all test ECGs were classified correctly. The best results were obtained using a learning tolerance of 2.5%. If the classification coronary artery disease vs. non-coronary artery disease was made using this trained neural network, 72% of all tachycardias were diagnosed correctly.<
>