Temporal EKG signal classification using neural networks

Abstract

A system is developed which makes diagnostic determinations from EKG signals utilizing Back Propagation Neural (BPN) Networks to window discretized temporal EKG signals and classify segmented signal waveform data. First, characteristics of EKG signal data are explored as they relate to the heart functions. Then, assuming a Gaussian behavior of EKG signals, a system is developed based on pattern matching of abnormal heart function characteristics. The windowing function of EKG signals is performed by using a neural network trained on apriori data. EKG signal feature extraction and waveform segmentation is performed and the results encoded for inputs to independent neural networks. These networks memorized signal data conditions to provide diagnostic scores that classify to within 4% of their initial case training weights. It is concluded that employing neural networks to perform temporal EKG signal classification is a viable, efficient approach.