Detection of atrial fibrillation using artificial neural networks

Artificial neural networks (ANNs) were used as pattern detectors to detect atrial fibrillation (AF) in the MIT-BIH arrythmia database. Electrocardiogram data were represented using generalized interval transition matrices, as in Markov model AF detectors (G.B. Moody and R.G. Mark, 1983). A training file was developed, using these transition matrices, for a backpropagation ANN. This file consisted of approximately 15 minutes each of AF and non-AF data. The ANN was successfully trained using these data. Three standard databases were used to test network performance. Post-processing of the ANN output yielded an AF sensitivity of 92.86% and an AF positive predictive accuracy of 92.34%