An efficient system for the detection of arrhythmic segments in ECG recordings based on non-linear features of the RR interval signal

In this paper we explore the RR interval signal to detect arrhythmic segments in electrocardiograms (ECG) using non-linear analysis. Initially, the RR interval signal is extracted and it is segmented into small segments. Linear (standard deviation), spectral (total energy) and non-linear (approximated entropy and normalized entropy) characteristics are extracted for each segment. Time-frequency analysis is used for the calculation of the total energy. These characteristics are fed into a neural network to classify each segment as normal or arrhythmic. The proposed approach is validated using the MIT-BIH database for various segment sizes (32, 64, 128, 256 and 512 RR intervals). The method results in high sensitivity and specificity (85% sensitivity and 92% specificity) for arrhythmic segment detection.