Segmentation of high-resolution ECGs using hidden Markov models

Accurate determination of the offset and onset of QRS complexes in a high-resolution electrocardiogram (ECG) is required for time-domain detection of late potentials. Late potentials are usually detected by applying a threshold algorithm to a signal-averaged and filtered QRS complex. Adaptive filtering algorithms have made it possible to attempt beat-to-beat analysis of late potentials but the standard threshold algorithm used for signal-average ECGs yield inconsistent endpoint detection. A new algorithm which uses hidden Markov modeling to represent the transitions between the QRS complex and surrounding isoelectric regions is shown to provide consistent QRS boundary detection despite variations in noise level in the range of 0-1 mu V. For 74 high-resolution ECGs analyzed, the mean standard deviation of the QRS offset point was reduced from 11 to 2 ms.<>