Session S34.5

Adaptive Threshold QRS Detector with Best Channel Selection Based on a Noise Rating System

F Chiarugi*, V Sakkalis, D Emmanoulidou, T Krontiris,
M Varanini, IG Tollis

Foundation for Research and Technology Hellas
Heraklion, Greece

QRS detection performance can depend on the type of noise present in each lead involved in the overall processing. Very often, when the number of leads is very low, a common approach to QRS detection is through the QRS enhancement achieved in a complex signal (CS) based on the derivatives of the pre-filtered available leads. However, the signal pre-filtering cannot be able to perform a complete noise rejection and the use of the derivatives can produce a noise enhancement as well. In many cases the noise occurs only on one lead and the addition of a noisy lead to the CS decreases the overall detection capabilities of the QRS detector. For this reason the noise evaluation on each channel, providing information for the inclusion or rejection of that lead in the building of the CS, can improve the overall performances of the QRS detector. The QRS detector is based on a pre-filtering of a two-lead ECG with a moving average linear filter in the 5-15 Hz band followed by a derivative filter applied on each channel. A CS is then generated summing the absolute value of each derivative signal. An adaptive threshold is then initially evaluated based on a percentage of the average of the CS peaks in the recording. After each first threshold crossing, the threshold is re-evaluated. A dead-time zone of 200 msec is applied in order to avoid false positive due to too close detections of QRS, and a method for increasing artificially the threshold till the end of the T-wave is applied for avoiding the false detection of T-waves as QRSs. After the QRS detection, the powers of signal and noise are estimated for each channel on the QRS and following TP interval, and the N/S ratio is then evaluated. A noise rating system is applied in order to decide if one of the two channels has to be rejected from the CS and, in that case, the QRS detector runs on the CS composed only by the other channel. The results have been evaluated on the 48 records of the MIT-BIH Arrhythmia Database where each ECG record is composed by 2 leads sampled at 360 Hz for a total duration of about 30 minutes. The annotated QRSs are 109494 in total. The results have been very satisfying on all the annotated QRSs with a specificity=99.57% and a sensitivity=99.71% higher than other studies based on the same database. Further investigations will be dedicated to the use of algorithms for the detection of ventricular fibrillation/flutter period (record 207) in order to avoid the wrong detection of fibrillation/flutter waves as QRSs, and to methods for a further optimization of the noise rating decision criteria.

(Abstract Control Number: 62)