Baseline Wander Removal and QRS Detection Using Holter ECG Signals

Cardiac arrhythmias (such as atrial fibrillation or ventricular tachycardia) may occur infrequently over the course of a day, month or even year, or may be induced by specific activities. Doctors often use an electrocardiograph (ECG) Holter tape to record either 24 or 48 hours of ECG signal, in order to detect the onset and occurrence of suspected arrhythmias. In order to diagnose a cardiac rhythm disorder, the ECG Holter tapes, which are lengthy and contain a substantial amount of data, must be analyzed quickly and accurately in an automated fashion. Biomedical Systems (St. Louis, MO), one such company that produces software necessary to automatically analyze ECG Holter tapes, would like to improve three aspects of their current software. The three improvements are: a filter to remove baseline wander, a new RPeak Filter and possibly an improved QRS detection scheme. Each of the tasks presented in the problem statement, baseline wander removal [6-12] and QRS detection [13-18] algorithms have been addressed previously. In the present commercial system, produced by Biomedical Systems (BMS), there is no method to remove baseline wander from the ECG signal. Previous techniques to remove baseline wander from an ECG signal include a four-pole, null phase digital filter [7], a two-pole, phase-compensated (PC) digital filter [8], a Finite Impulse Response filter [9, 10], a sixth-order polynomial [11] and a Cubic Spline Technique [6, 12]. The cubic spline technique, a nonlinear, adaptive filter, depends upon a reliable method of estimating baseline points. It is assumed the best estimator is a point just before the QRS onset ( optimal local fit), but this assumption may break down in some circumstances, such as for premature ventricular contraction (PVC) on the T-wave. Furthermore, detection of QRS complexes may be confounded by abrupt shifts in the baseline [7]. The sixth-order polynomial is based upon a global best-fit method. In small data sets, the sixth-order polynomial performed superior to the cubic spline method, however the authors did not compare the two for long data sets. The four-pole, null-phase filter works by applying a filter with a cut-off frequency of 0.60 hertz to the data, first in the forward direction, and then in the reverse direction to remove any non-linear phase components introduced by the filter. The two-pole, phase-compensated digital filter, based upon the four-pole, null phase filter, allows for quicker processing of the data. The primary strength of the null-phase filter is the ability to reduce baseline wander with minimum distortion to the ST-segment. The finite impulse response (FIR) filters apply a high pass filter to the data, to remove frequencies below 0.5 hertz. The primary disadvantages of FIR filters are the lack of conformity to the AHA frequency response requirements. That is, FIR filters fail to reduce abrupt shifts in the baseline and may introduce distortion in the ST-segment. The null-phase digital filter, phase-compensated digital filter, cubic spline technique and sixth-order polynomial all satisfy AHA frequency response requirements..