06-11-2012, 03:22 PM
Baseline Wander Removal from Pulse Signal
1Baseline Wander.pdf (Size: 336 KB / Downloads: 31)
Abstract
The problem of efficient removal of the baseline wandering from
the pulse signal has been an area of focus over the past few years.
The baseline wander is the consequence of the artifacts introduced
due to respiration during the pulse waveform acquisition. The
objective here is to propose a new algorithm to detect the baseline
wander and to use the same in a wavelet based cascaded adaptive
filter (CAF) for removing the baseline wander noise. This filter
works in two stages, in which the initial filtering is done by a
Discrete Meyer filter followed by Cubic Spline Estimation. Since the
frequency content of the baseline wander signal spreads from 0.1 to
0.8 Hz, the incoming signal is passed through two blocks. The initial
wavelet filter is efficient for the removal of the frequency components
0.4 to 0.8 Hz, while the new proposed algorithm (Baseline
Estimation block) focuses on the much lower frequency content 0.1 to
0.4 Hz. Finally the results are compared with the traditional method,
which shows that the proposed algorithm has a better performance
for removing the baseline wander.
INTRODUCTION
In the field of medical sciences, various physiological
signals like ECG, radial pulse signal are used for the diagnosis.
With the view for automation of diagnosis, these signals are
required to be processed in different manners. The inherent
noise, present in the patient‟s physiological signal, poses a
major problem in automation, leading to misdiagnosis. So it
is essential not only to remove the noise but also to maintain
the original morphology of the signal undisturbed. Baseline
wander is one such type of noise that arises due to respiration
and motion artefacts. Here we discuss a new method to
remove baseline wander from the radial pulse signal.
Signal Decomposition:
To decompose the pulse signal using
wavelet technique and compute its ER to detect
pulse‟s baseline wander level. When the baseline
wander is high, the wavelet filter performs better
than cubic spline estimation. When there is little
baseline wander, the wavelet filter may introduce
some distortion. However, it is not the case when
both the signal and the baseline wander are low. It is
observed that, in this case, the wavelet filtering
performs better than the spline estimation filter.
Basically the performance of the wavelet filter does
not depend on the absolute magnitude of the noise
only but it depends on the relative magnitude of the
noise with that of the pulse signal.
Energy Ratio (ER) is the figure which roughly
indicates the ratio of the energy of the pulse signal to
that of its baseline wander. Thus, when the ER of a
pulse waveform to its baseline wander is sufficiently
high, relative energy of the baseline wander is small,
and so spline estimation corrects baseline wander
more effectively than a wavelet filter. On the other
hand, when the ER of a pulse waveform to its
baseline wander is low, wavelet filter corrects
baseline wander more effectively than spline
estimation.
Discrete Wavelet Filter:
Filter out the low frequency noise signal (particularly in the band 0.4Hz to 0.8Hz) using discrete wavelet filtering (Here, Discrete Meyer Wavelet is used for this purpose). If the value of ER is less than a certain empirically fixed threshold, the wavelet filtering is done. In this case, the highest level of the approximation the physiological signal (i.e. the baselines wander) is subtracted from the raw signal. This process enhances ER of the signal. Signal thus filtered from the wavelet filter is passed further to next stage of spline estimation filter. If ER is already greater than the threshold, the raw signal is directly passed to the estimation filter stage.
Cubic Spline Estimation:
Determine the onsets of the pulse signal with help of an algorithm, based on the pulse waveform‟s amplitude and derivatives. The onset points are used as the knots for the spline estimation. To remove the remaining contamination, we detect the „reference points‟ of the pulse1 and apply the cubic spline to correct the baseline and normalized the pulse waveform. Some researchers apply the linear interpolation method. But the wander caused by the respiration and motion. The waveform of respiration is nonlinear and quasi-periodic. Thus the linear interpolation method must cause some distortion. Table1 illustrates that when the signal to noise ratio is high the result of cubic spline interpolation is satisfied. According to the pulse waveform complex, the onset points can be detected very accurately. Having been filtered by cubic spline estimation, the pulse signal is normalized and easy to be computed and analysed.
EXISTING METHOD [1]
In this method, the experiments are done on both types of the signal; viz. the raw real pulse data and the simulated pulse data. The real data used is the actually recorded pulse signals of number of patients. The simulated data is the pulse signal generated by adding the noise signal in the ideal expected signal waveform. The low frequency periodic components are added to simulate the baseline wander resulting from respiration. The nonperiodic components of pulse baseline wander which were simulated by the low-pass filtered random noise, represent the aperiodic motion artifact.
RESULTS: DISCUSSION AND COMPARISON
Here, to compare the performance of the proposed algorithm
with that of the existing one, BCR (Baseline Correction Ratio)
for the two methods is evaluated and compared. BCR is a
quantitative measure of the extent to which the baseline
wander is removed. The BCR is determined as [1]