|
Digital Signal
Processing Methods for the Evaluation of Blood Volume Pulse (BVP)
Waveform Changes due to Exercise
Chao
Li, Jing Zhai and Armando Barreto
I. Introduction
Finger photoplethysmography
(PPG) is a non-invasive monitoring technique, which does not
require costly equipment or specialized personnel.
Traditionally, the Blood Volume Pulse (BVP) has been used to
determine the heart rate [1]. However a more detailed analysis
of photoplethysmographic BVP variations may indicate circulatory
changes that take place in an individual due to exercise.
Further research has shown that the changes in the single BVP
signal through an exercise session (before exercise, immediately
after exercise, and after a recovery period) can be measured,
characterized, and quantified through signal processing methods
[2].
II.
Data Gathering
Protocol
The recording procedure used
consisted of three measurements: A, B and C, through a total
time of about 18 minutes. Ten healthy volunteers with ages
between 22 and 40 participated in the recordings.
Each subject was requested
to sit comfortably, resting the right arm on a table. At this
point, the first photoplethysomgraphic Blood Volume Pulse record
was obtained from the tip of the right index finger (Stage “A”,
“Before exercise”). After recording for about 30 to 60 seconds,
the subject was asked to begin performing continuous lifting of
a 6.6-pound dumbbell with the left arm. At the end of 1-minute
intervals, the subject was asked to switch the arm used to lift
the dumbbell. This process lasted for 8 minutes. Then a second
BVP measurement was taken (Stage “B”, “Immediately after
exercise”). After this, the subject was asked to rest, allowing
for his or her cardiovascular system to recover, for 8 more
minutes. After the recovery period, one last BVP measurement was
taken from the right index finger (Stage “C” or “After
recovery”).
Data Gathering Protocol:
The following three figures
show the BVP waveforms recorded from one of the subjects (S3) at
stages A, B and C, respectively. As we can see, in stages A and
C, the dicrotic notch is very pronounced, while in stage B the
Dicrotic Notch is less obvious, because of the changes in the
cardiovascular system caused by the exercise. This is probably
due to a combined effect of the increased cardiac output and
decreased peripheral resistance associated with exercise [4].
Our objective is to find methods to derive a single numerical
parameter from the BVP signal, which will reflect those
cardiovascular changes.
FIG1. BVP waveform from Subject S3 at stage A
FIG2. BVP waveform from Subject S3 at
stage B
FIG3. BVP waveform from Subject S3 at
stage C
III. Data processing methods
Three approaches were tested to quantify the changes introduced
in the photoplethysmographic Blood Volume Pulse waveform by
exercise.
1. Average Beat Histogram Analysis
Preliminary observation of
the records obtained for the study indicated that the dicrotic
notch tends to be less pronounced in the measurement taken right
after the exercise (Stage B). This led us to implement a signal
processing scheme that separates individual beats in the signal,
aligns them according to their maxima and computes a normalized
histogram for each of them. In this histogram the full amplitude
range of each beat is divided into 100 bins and the number of
values within each amplitude bin is found. Before the exercise
session (A) and after the recovery period (C), the presence of
dicrotic notch results in a large number of samples detected in
the range R1: (40% - 70%) of the total amplitude of the beats,
which tends to predominate over R2: (70% - 100%). On the other
hand, immediately after the exercise session the Dicrotic Notch
is normally less pronounced. So the ratio of the histogram
accumulation for range R1 and the histogram accumulation for R2
was selected as the single index to summarize the changes in the
histogram [3]. The following table indicates the ratios found
for the ten subjects at the 3 different stages of the procedure,
normalized to the value found in stage A:
|
|
A |
B |
C |
|
S1 |
1.00 |
0.7294 |
0.9132 |
|
S2 |
1.00 |
0.3839 |
0.9089 |
|
S3 |
1.00 |
0.2013 |
1.0873 |
|
S4 |
1.00 |
0.5365 |
1.0215 |
|
S5 |
1.00 |
0.7838 |
0.7962 |
|
S6 |
1.00 |
0.8422 |
0.8879 |
|
S7 |
1.00 |
0.4042 |
0.5712 |
|
S8 |
1.00 |
0.9264 |
0.9182 |
|
S9 |
1.00 |
1.1282 |
1.3340 |
|
S10 |
1.00 |
0.5276 |
0.8639 |
Table I.
Histogram method results
FIG 4
Histogram method results
2.
BVP
Harmonic Composition Changes
The presence of the
Dicrotic Notch in the BVP waveform is represented in the
frequency domain as a significant component corresponding to the
second harmonic of the fundamental BVP frequency. This second
harmonic component is normally very significant in comparison
with the fundamental for stages A and C (i.e., at rest or after
recovery). In contrast, the second harmonic contribution is
small at stage B (right after the exercise session), since the
Dicrotic Notch is less well defined. To detect these changes we
propose the determination of the ratio Pf2/Pf1, where Pf2 is the
amplitude of the second harmonic and Pf1 is the amplitude of
fundamental in the averaged periodogram of the BVP segment
recorded at each stage of the procedure. The values of this
ratio, normalized to the ratio measured in stage A are shown in
Table II and Figure 3.
|
|
A |
B |
C |
|
S1 |
1.00 |
1.2965 |
0.9408 |
|
S2 |
1.00 |
1.1722 |
0.8228 |
|
S3 |
1.00 |
1.6172 |
0.9418 |
|
S4 |
1.00 |
1.0162 |
1.0092 |
|
S5 |
1.00 |
0.6991 |
0.8178 |
|
S6 |
1.00 |
0.9147 |
1.6994 |
|
S7 |
1.00 |
1.4692 |
1.3159 |
|
S8 |
1.00 |
1.1322 |
1.4105 |
|
S9 |
1.00 |
1.9591 |
1.7746 |
|
S10 |
1.00 |
1.3119 |
0.7448 |
Table II
Harmonic method result
FIG 5
Harmonic method results
3. Dicrotic
Notch depth method:
The algorithm for this
approach isolates each individual BVP beat and identifies the
local minimum and local maximum associated with the presence of
the Dicrotic Notch. The difference between these two levels is
used as the estimation of the Dicrotic Notch depth for a
particular beat. The ratio of the Dicrotic Notch depth to the
total amplitude of each beat is found and an average of this
ratio is calculated for all the beats in the BVP segment under
study. Table III and Figure 4 show the results for the 10
subjects studied, at the 3 stages of the protocol, normalized to
make stage A for every subject 1.0:
|
|
A |
B |
C |
|
S1 |
1.00 |
0 |
0.0984 |
|
S2 |
1.00 |
0.0550 |
0.4990 |
|
S3 |
1.00 |
0.0084 |
2.3933 |
|
S4 |
1.00 |
0.0351 |
1.0879 |
|
S5 |
1.00 |
0.0905 |
0.2032 |
|
S6 |
1.00 |
0.0263 |
0.0088 |
|
S7 |
1.00 |
1.00 |
1.00 |
|
S8 |
1.00 |
1.00 |
1.00 |
|
S9 |
1.00 |
0.0056 |
0.5506 |
|
S10 |
1.00 |
0.4069 |
1.1176 |
Table
III
Dicotic notch depth method results
FIG 6
Dicrotic notch depth method results
IV. Discussion
Tables I – III and Figures 4- 6
show that there is significant variability in the BVP responses
recorded from different individuals through the exercise
protocol used in this study. Figure 5 shows that the evolution
of the harmonic ratio measure was particularly different for
different individuals. It is important to assess if these
differences are primarily due to the methods used to evaluate
the changes in BVP waveforms or whether they arise because the
waveforms themselves varied in different ways through the 3
recorded stages (A, B and C) for different individuals. By
visual inspection of the waveforms recorded, we realized that
for eight of the ten experimental subjects the waveforms at the
different stages showed the anticipated changes very clearly.
However, the waveforms from two of our subjects (S8 and S9)
showed much less pronounced changes.
In order to compare the
efficiency of each one of the signal processing approaches used
in the study to represent the changes in the BVP waveform
induced by exercise, we will focus on the records obtained from
the eight subjects for whom BVP waveform changes were clear,
i.e. we will exclude from consideration the results from
subjects S8 and S9. Looking at this subset of the data we
appreciate that the Histogram method (Table I and Figure 4)
provides results that are consistent with our expectations. For
this method the lessening of the Dicrotic Notch observed
immediately after exercise (stage B), is represented by a
reduction of the numerical index derived through the methods,
with respect to the values obtained at stages A and C. For the
Histogram method this pattern is only disrupted for subjects S8,
which has a lower value for stage C than for stage B and for S9
whose value for stage A is lower than for stage B. These are,
however, two subjects with non-representative BVP waveforms.
For the Dicrotic Notch Depth
method, it encountered some problems when the waveform shows no
notch in the downward slope, but just a lessoning of the slope.
In such cases, the dicrotic notch depth is zero, which is the
case for S7 and S8. The indexes for S7 and S8 for the three
stages are all zero.
For the harmonic ratio method,
it indicate a lack of consistency of this measure, even within
the set of eight subjects whose BVP waveforms appeared typical
according to visual inspection.
V. Conclusion
Photoplethysmographic BVP
records obtained from 10 healthy volunteers confirm that the BVP
waveforms change as the subject performs exercise. In our study
eight of the 10 subjects studied displayed clear BVP waveform
changes of the type that we had originally expected. In the
remaining 2 subjects the BVP changes were subtler. The Histogram
method was successful in consistently representing the BVP
changes of the eight subjects whose data was typical. The
results from the Harmonic Ratio method were the least consistent
of the three methods considered. Overall, the Histogram seems to
be the most promising for the intended application, due to the
consistency of the results obtained and its relatively low
computational cost.
Reference:
[1] Lee A. L., Tahmoush A.
J., and Jennings J. R., “An LED transistor photoplethysmograph”,
IEEE Transactions on Biomedical Engineering, May 1995, pp.
248-250.
[2] Heimer M. and Barreto
A., “Evaluation of physical exercise using photoplethysmography”.
Proc. X Annual Conf. IEEE Engineering in Medicine and Biology
Society, November 1988, pp. 1617-1618.
[3] Barreto A., Heimer M.,
and Garcia M., “Characterization of Photoplehtysmographic Blood
Volume Pulse Waveforms for Exercise Evalution,” Proceedings 14th
Southern Biomedical Engineering Conference, Shreveport,
Louisiana, April, 1995, pp. 220-223.
[4] McArdle W., Katch F.,
and Katch V., Exercise Physiology: Energy, Nutrition and Human
Performance. Lea & Febiger, Philadelphia, 1985.
|
