Vous êtes sur la page 1sur 16

Human Factors and Ergonomics in Manufacturing, Vol.

19 (5) 478–493 (2009)



C 2009 Wiley Periodicals, Inc.

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hfm.20146

Multiscale Entropy Approach to Physiological Fatigue


during Long-Term Web Browsing
C. H. Hung
Department of Banking and Finance, Takming University of Science and
Technology, Taipei, Taiwan, ROC
Bernard C. Jiang
Department of Industrial Engineering and Management, Yuan Ze University,
Taoyuan, Taiwan, ROC

ABSTRACT

Physiological fatigue during long-term Web browsing was investigated by the entropy-based method.
A noninvasive instrument was used to gather 8-hour continuous electrocardiography signals from
subjects who were asked to answer the psychological fatigue questionnaire every 2 hours. These
data were used to calculate the heart rate (HR), traditional sample entropy (SampEn), and multiscale
entropy (MSE). HR decreased as the browsing task began, but increased slightly after 4.5 hours. The
psychological fatigue score gradually increased, implying that more severe subjective fatigue was
experienced as Web browsing proceeded. In contrast, the traditional SampEn first increased and then
oscillated after 2.5 hours. The browsing time of 2.5 hours may be a clue to physiological fatigue.
The MSE results showed that the cardiac dynamic systems of undergraduates and males were more
complex than those of graduates and females, respectively. Thus, SampEn may have the potential for
estimating physiological fatigue during long-term Web browsing.  C 2009 Wiley Periodicals, Inc.

1. INTRODUCTION
With the rapid developments in computer technology and the widespread nature of the
Internet, the number of Internet users is increasing year by year. Internet activities on the
World Wide Web (WWW) are becoming progressively more important in daily life due
to their manifold applications (e-business, e-mail, Web browsing, online games, etc.). As
was reported in 2007 (Taiwan Network Information Center [TWNIC], 2007), the three
most popular activities for Internet users in Taiwan are Web browsing (67.37%), e-mail
usage (40.34%) and information searching (26.44%). Moreover, at least 25% of Internet
users spent more than 4 hours online daily (TWNIC, 2007). Long-term computer users
may complain of adverse psychological and physiological symptoms. Furthermore, recent
research (Beasley, Raymond, Hill, Nowitz, & Hughes, 2003) has shown that sitting at a
computer terminal for prolonged periods may result in vein thromboembolism.
Uncomfortable symptoms arising from video display terminal (VDT) tasks may include
visual fatigue (Bergqvist & Knave, 1994; Sheedy, 1992) and musculoskeletal complaints (Li

Correspondence to: C. H. Hung, Department of Banking and Finance, Takming University of Science and
Technology, 11451, Taipei, Taiwan, ROC. E-mail: chhung@mail.takming.edu.tw

478
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 479

& Buckle, 1999), and much research on VDT tasks has been reported. Questionnaires have
been widely used to investigate psychological fatigue during VDT tasks (Dillon & Emurian,
1995; Dillon, Kleinman, Choi, & Bias, 2005; Duffy & Chan, 2002; Stüdeli & Menozzi,
2003), whereas other studies have gathered physiological signals by electroencephalogra-
phy (Shieh, Chen, & Wang, 2005), electromyography (Balci & Aghazadeh, 2004; Murata,
Uetake, Matsumoto, & Takasawa, 2003; Park, Kim, & Shin, 2000; Seghers, Jochen, &
Spaepen, 2003), and critical flicker fusion (Chi & Lin, 1998; Takahashi et al., 2001) to mea-
sure physiological fatigue. Heart rate (HR) has been measured in VDT tasks at low load levels
(Higuchi, Motohashi, Liu, Ahara, & Kaneko, 2003; Takahashi et al., 2001); however, the
change in HR is not remarkable when it is compared with the high loading task like HR recov-
ery after exhausted exercise (Du et al., 2005; Platisa, Mazic, Nestorovic, & Gal, 2008). Even
for the same HR, the measures of HR variability (HRV) are different (Platisa & Gal, 2006).
Because it is well known that HR is not static, the variation in the interval between
consecutive heartbeats, called HRV, contains more information than the mean HR. HRV may
have indicators of current disease or impending cardiac diseases and is a reliable reflection
of the physiological factors modulating the normal rhythm of the heart. It is also thought
to reflect the heart’s ability to adapt to changing circumstances by detecting and quickly
responding to unpredictable stimuli (Acharya, Joseph, Kannathal, Lim, & Suri, 2006).
HRV is used as a clinical tool to diagnose cardiac autonomic function in both health and
disease (Kim, Yoon, Bai, & Soh, 2005). Electrocardiography (ECG) data, measured easily
and noninvasively, contain the time series of beat-to-beat intervals which can be analyzed
to calculate HRV. There are roughly three classified measures for HRV. The first are time-
domain measures, which are simple and usually have statistical parameters (e.g., the standard
errors, standard deviation (SD), and root-mean-square values of the samples from R-R
intervals [RRIs] in a 0.5- to 5-minute time window) as indices. Second, frequency-domain
measures can distinguish between sympathetic and parasympathetic contributions of HRV
by using the periodogram or fast Fourier transform for the power spectral density estimation
procedure. And third, because it has been shown that the structure generating biologic
signals is nonlinear (Akay, 2000), nonlinear domain measures such as largest Lyapunov
exponents (Wolf, Swift, Swinney, & Vastano, 1985), detrended fluctuation analysis (Peng
et al., 1995), approximate entropy (ApEn; Pincus, 1991), sample entropy (SampEn; Richman
& Moorman, 2000), and so forth, have been used to describe nonlinear fluctuations in a
more efficient manner.
Because biologic systems have to adapt to complex environments, the evolution process
makes all biologic systems extremely complex. However, there is no formal definition of
complexity. Intuitively, complexity is related to our ability to provide a short description
of a phenomenon. The mathematical definition of complexity follows from information
theory; however, with the exception of some theoretical applications, the mathematical
complexity of a system can not be easily calculated. For practical applications, several
entropy-based methods, such as ApEn and SampEn, which are nonlinear measures of HRV,
have been proposed to measure complexity (Costa, Peng, Goldberger, & Hausdorff, 2003).
The important application of entropy lies in the fact that it is a measure of the disorder
(complexity) in the HR signal (Acharya et al., 2006).
SampEn, statistically described by Richman and Moorman (2000), is a more refined
form of ApEn and has been used to quantify complexity (irregularity) of RRI time series,
or how chaotic the dynamic system is. It is defined as the negative natural logarithm of
the conditional probability that two sequences similar for m points remain similar within
tolerance r at the next point, where self-matches are excluded. In other words, it quantifies
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
480 HUNG AND JIANG

the extent to which a sequence of m RRIs can predict the next RRI duration, based on the
knowledge of the degree of the similarity for sequences of length m to that for sequences of
length m + 1 (Vuksanovié & Gal, 2007). A time series containing many repetitive sequences
has a relatively small SampEn, and a less predictable process has a higher SampEn. Thus,
a high value of SampEn reflects a low degree of regularity, whereas a lower value of
SampEn indicates more self-similarity in the time series. Because SampEn is a powerful
tool of the HRV measures, it has many applications. Gour et al. (2007) used SampEn to
measure the regularity of the movement of the patients with Parkinson’s disease. SampEn
was successfully used to differentiate fetal cardiac signals from maternal and noise ones
(Comani, Srinivasan, Alleva, & Romani, 2007). The research of Abásolo, Hornero, Espino,
Álvarez, and Poza (2006) suggests that SampEn could yield essential information and may
contribute to increasing the insight into brain dysfunction in Alzheimer’s disease in ways
which are not possible with more classical and conventional statistical methods.
Traditional entropy-based algorithms measure the degree of regularity (orderliness) of a
time series by the occurrence of repetitive patterns. Entropy increases with the degree of
disorder and reaches a maximum for completely random systems. Based on the fact that the
entropy of a time series depends on its SD and correlation properties, uncorrelated stochastic
signals with larger variances have higher entropy values, whereas the entropy of a periodic
signal is smaller than that of a random one (Costa, Goldberger, & Peng, 2005).
Because pathologic time series represent the output of less adaptive (complex) systems,
entropy values are presumably higher for normal subjects and smaller for cardiac abnormal
subjects (Acharya et al., 2006). Nevertheless, the statement that pathologic time series rep-
resent the output of less adaptive and less complex systems (Mandell & Shlesinger, 1990;
Paulus et al., 1990) is not always true. For example, atrial fibrillation (AF), associated with
highly erratic fluctuations with statistical properties like uncorrelated white noise (Balocchi
et al., 1999; Hayano et al., 1997; Zeng & Glass, 1996), yields higher entropy values as
compared to healthy dynamics showing correlated (1/f-type) properties, even though the
latter represent more complex and adaptive states (Costa, Goldberger, & Peng, 2002). Costa
et al. (2002) generated the (1/f) noise as follows: They calculated the fast Fourier transform
(FFT) from uniformly distributed white noise, and then calculated the inverse FFT with
imposing a (1/f) distribution on the power spectrum. Because the time series of physiolog-
ical systems, regulated by interacting mechanisms that operate across multiple spatial and
temporal scales, contain deterministic and stochastic structures, the misleading result has
been addressed by introducing the multiscale entropy (MSE) method (Costa et al., 2005).
Costa et al. (2005) used the MSE method for quantifying the complexity expressed by
physiological dynamics over multiple scales (from scales 1 to 20). They first applied the
MSE method to uncorrelated white and correlated 1/f noises. For scale 1 (the original
time series), the entropy derived from the white noise is larger than that from 1/f noise.
As the scale increases (coarse-grained time series), the entropy derived from the white
noise monotonically decreases, whereas the entropy derived from 1/f noise remains almost
constant. For scales greater than 5, the entropy for white noise becomes smaller than that
for 1/f noise. The result indicates that 1/f noise contains complex structures across multiple
time scales and that white noise degrades for large scales. They further compared the
physiological signals from healthy subjects, patients with severe congestive heart failure
(CHF), and patients with cardiac arrhythmia, AF. Three results were stated in their research:
(1) The entropy value derived from healthy subjects increases at small scales and then
stabilizes to a constant value. (2) The entropy value derived from patients with CHF markedly
decreases at small scales and then gradually increases. (3) The entropy value derived from
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 481

patients with AF monotonically decreases, similar to white noise. For small scales, the
AF patients have higher entropy values than do the CHF patients and healthy subjects. As
the scale increases to a sufficiently large value, the entropy values of healthy subjects are
the highest, indicating the most complex signals for healthy subjects. They suggested that
both the specific values of the entropy measure and their dependence on scales need to be
considered in the physiological process. Costa et al. (2005) also applied the MSE method
to distinguish the complexity of the waking period from that of the sleeping period, and
healthy young subjects from healthy elderly subjects.
Although many work–rest schedules for VDT tasks (Balci & Aghazadeh, 2004; Bouc
& Thum, 1997) have been suggested to reduce physiological and psychological fatigues,
Internet users may violate these advices and simply continue browsing because of the
attractive nature of the content on the WWW. Furthermore, no research on entropy-based
measures to quantify the complexity of ECG time series of subjects during long-term Web
browsing has been reported. In this study, ECG signals were gathered continuously from
subjects engaged in 8-hour browsing tasks, and the subjects were also asked to answer
psychological fatigue questionnaires every 2 hours. After that, SampEn was adopted to
measure the complexity of cardiac dynamics (biologic systems) of subjects with respect to
browsing time and the results were also compared with the psychological fatigue scores.
The MSE measures were used to uncover the scale-dependence of the physiological systems
and explore the complexity of cardiac dynamics (biologic systems) among different groups
of subjects during browsing tasks.

2. METHOD
In this section, we describe the experimental and analytical methods used in the study.
The subjects who participated in the experiment, the experimental setup, the experimental
procedure, the methods of data collection, and data analysis are described in the following
subsections.

2.1. Subjects
Fifteen volunteer students were used as the subjects in the experiment. Seven were under-
graduates (three males and four females) of age 19 ± 1 years, and the others were graduates
of age 23 ± 1 (four males and four females). These young subjects were all right-handed
and accustomed to browsing the WWW with Internet Explorer for more than 2 hours a day.
They were clear-sighted and not color blind; and they were healthy nonsmokers who did
not report chronic pain symptoms in the head, neck, shoulder, or back. They were asked to
abstain from medicines, coffee, or alcoholic beverages before the experiment.

2.2. Experiment Protocol


The indoor noise index was controlled to less than 50 dB, and the illuminance level was
maintained at 750 lux. The desk height was 67 cm, and the chair height was 50 cm. The
distance between the subject and monitor was 60 ± 5 cm, and the viewing angle was 20 ±
5◦ . Room temperature was kept between 18◦ C and 22◦ C.
The experimental instruments included a desktop computer with a 17-inch Thin Film
Transistor Liquid Crystal Display (TFT-LCD) monitor and a Philips IntelliVue MP60 patient
monitor (made in Germany). The desktop computer was equipped with an Intel Pentium 4,
2.8-GHz CPU with 1 GB of RAM. The subjects used Internet Explorer 6.0 to browse
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
482 HUNG AND JIANG

the WWW via Taiwan Academic Network (TANET) on the Microsoft Windows XP SP2
operating system.
Almost 400 similarly structured Web sites, including those of many departments of
universities and the government, were selected and listed on the homepage as the subjects
might have found it easier to click on the links to these Web sites from the homepage.

2.3. Procedures
The subjects participated in the experiment individually, and it took 15 days (15 subjects)
to complete the experiment. The experiment was performed from approximately 8:00 am
to 4:00 pm. The subjects arrived at the room before 7:30 am to prepare for the experiment.
When the subject entered the room, he or she was asked to sit in a comfortable posture
for 30 minutes. Subsequently, he or she completed the psychological fatigue questionnaire
and then practiced five demo questions the answers of which could be found on the Web.
The probes of the physiological monitoring system (Philips IntelliVue MP60) were fixed
on the subject, and the instrument started to record ECG signals just before the experiment
began. When the experiment began, the browsing time was simultaneously recorded. During
the 8-hour experiment, the subject was randomly assigned to browse different Web sites.
The subject was given and asked to complete the psychological fatigue questionnaire every
2 hours. He or she was not allowed to rest, talk, drink, or drastically change his (her) posture.

2.4. Data Collection


2.4.1. Psychological Fatigue Questionnaire. The psychological fatigue question-
naire, extracted from the report of Mao et al. (1998), included 15 fatigue symptoms, each
of which had 7 score levels. The minimum score of 0 was recorded if the subject absolutely
disagreed with the description, whereas the maximum score of 6 was recorded if the subject
absolutely agreed with the description. The 15 symptoms were described in the form of
questions asking if the subject would like to (1) yawn, (2) lie down, or (3) go to sleep;
experienced (4) eye fatigue, (5) shoulder ache, (6) waist ache, (7) eyelid twitching or (8)
hand ache; lost the ability to (9) think clearly, (10) concentrate or (11) induce; (12) ran out
of patience; (13) was in a bad mood; (14) was thirsty; or (15) felt dizzy.

2.4.2. ECG Biologic Signals. The Philips IntelliVue MP60 patient monitor was used
to gather continuous ECG data from the subjects at a sampling rate of 500 Hz. The desktop
computer was connected to the patient monitor so that the ECG data could be transferred and
stored simultaneously. Therefore, information on the RRIs could be obtained to calculate
the HR and SampEn.

2.5. Data Analysis of SampEn and MSE


SampEn was developed by Richman and Moorman (2000); it can be calculated as follows:
Let {Xi } = {x1 , . . . , xi , . . . , xN } represent a time series of length N , and consider the
m-length vectors um (i) = {xi , xi+1 , . . . , xi+m−1 }, 1 ≤ i ≤ N − m. Let nm i (r) represent the
number of vectors um (j, j = i) that are close to the vector um (i) (i.e., the number of vectors
that satisfy d[um (i), um (j )] ≤ r, where d[um (i), um (j )] = max{|x(j + k) − x(i + k)| : 0 ≤
k ≤ m − 1}). The tolerance level r is set at a certain percentage of the SD ofthe time series.
Finally, SampEn can be calculated by SampEn (m, r, N ) = ln( N−m m N−m m+1
i=1 ni / i=1 ni ).
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 483

The MSE method is used to reconstruct a consecutive coarse-grained time series


from an original time series by averaging the data in nonoverlapping windows. Let
{Xi } = {x1 , . . . , xi , . . . , xN } represent an original time series of length N
. Then, each ele-

ment of a coarse-grained time series, yjτ , is calculated by the equation yjτ = i=(j −1)τ +1 xi /τ ,
1 ≤ j ≤ N/τ , where τ is the scale factor. The scale is small if τ < 5, and large if τ > 5. We
can use the coarse-grained time series generated at various scales to compute SampEn. It
should be noted that r, a certain percentage of the original time-series SD, remains constant
for all scales in the MSE method. For scale 1 (τ = 1), the reconstructed time series is simply
the original time series and the result is the traditional entropy-based measure.
SampEn is a refinement of ApEn and it can measure the complexity of signals. It is largely
independent of N when the total number of data points is larger than approximately 750
(Costa et al., 2003). For the MSE method, the suggested minimum number of data points
for the shortest coarse-grained time series is 1,000 (Costa et al., 2005).
In this research, we use m = 2 and r = 0.15 SD, and adopt consecutive RRIs of a certain
time period as the time series (Costa et al., 2005). Because a 30-minute time series normally
contains 1500 to 2400 RRIs during the browsing task, we calculate the traditional SampEn
(τ = 1) for the data of the 30-minute time interval. For the MSE method, we calculate the
SampEn at scales 1 to 12 for 4-hour intervals and at scales 1 to 20 for 8-hour intervals.
The ECG data and the psychological fatigue scores are analyzed and discussed in the
next section.

3. RESULT AND DISCUSSION


The subjects who completed the 8-hour continuous Web-browsing tasks looked tired and
uncomfortable. After collecting the physiological ECG data and the psychological fatigue
scores from each subject, we calculated the HR, psychological fatigue score, traditional
SampEn, and MSE. The results accompanied by discussions are presented below.

3.1. HR
The 8-hour ECG signals were uniformly subdivided into 16 sequential parts so that each
contained consecutive RRIs of a 0.5-hour time series. Subsequently, the mean HR values of
subjects were calculated; the result is shown in Figure 1, in which HR is plotted at the end
of each time interval on the horizontal axis. The mean HR is approximately 76 beats per
minute (bpm) during the first half hour of the browsing task, and it monotonically decreases
with browsing time. This decreasing trend agrees with the findings of previous VDT studies
(Higuchi et al., 2003; Takahashi et al., 2001). The mean HR reaches a minimum of 68 bpm
at 4 to 4.5 hours. As the browsing task continues, the mean HR increases slightly and finally
approaches 70 bpm. We observed that the subjects looked uneasy, changed their postures,
and breathed deeply more frequently as the browsing task proceeded for a long time. These
behaviors may have caused a small increase in HR late in the 8-hour period.

3.2. Psychological Fatigue Score


The psychological fatigue scores, reflecting the subjective feeling of fatigue symptoms,
were recorded at browsing times of 0 (just before browsing), 2, 4, 6, and 8 hours (just after
browsing). The scores of the subjects were averaged, and the result is plotted in Figure 2.
The curve illustrates that the mean psychological fatigue score monotonically increases
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
484 HUNG AND JIANG

Figure 1 Heart rate against the browsing time (solid circle = mean, vertical bar = standard error).

Figure 2 The mean psychological fatigue score at the browsing time of 0 hr, 2 hr, 4 hr, 6 hr, and
8 hr.

from 3 to 29 during the browsing task. By using the correlated sample analysis of variance
(ANOVA), the hypothesis that the mean scores recorded at the five browsing times are all
the same is rejected (Mauchly’s sphericity test p < 0.05; Greenhouse–Geisser test p <
0.05). Furthermore, the result of the least significant difference method shows that the mean
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 485

Figure 3 Traditional SampEn against the browsing time (solid circle = mean, vertical bar =
standard error).

scores recorded at the browsing times of 2 and 4 hours are not significantly different, but
they are significantly different at 0, 2 (or 4), 6, and 8 hours.

3.3. Traditional SampEn


The 8-hour ECG signals arranged for HR calculations are also used to estimate the traditional
SampEn values. The mean SampEn is plotted in Figure 3. The lowest SampEn (1.20) occurs
at 0 to 0.5 hour, whereas the highest value (1.45) occurs at 6 to 6.5 hours. The plot in Figure 3
shows that SampEn increases after Web browsing begins, and then oscillates after 2.5 hours.
The first slight decrease (0.013) at 2.5 to 3 hours may be a clue of the physiological fatigue.
If at this point the subject had stopped browsing and rested for a while, he or she might
have recovered from discomfort (Boucsein & Thum, 1997). As the browsing continues
without resting, the second (0.030) and third (0.056) decreases, which come after relative
higher values of SampEn, occur at 4.5 to 5.5 hours and 6.5 to 7.5 hours. The decrease in
SampEn may originate from the self-adjustable mechanisms of the biologic system and
may somewhat reflect the extent of physiological fatigue. Therefore, the highest SampEn,
recorded during from 6 to 6.5 hours, may imply the largest extent of physiological fatigue
during the 8-hour browsing task. By using the one-way analysis of variance (ANOVA), the
hypothesis that the mean SampEn values at 0 to 2 hours, 2 to 6 hours, and 6 to 8 hours are all
the same is rejected (F test p < 0.05). The result of the least significant difference method
shows that the entropy values at 0 to 2 hours are different from those at 2 to 6 hours and 6 to
8 hours, but the entropy values at 2 to 6 hours and 6 to 8 hours are not significantly different.
The result implies that the states of cardiac dynamics are different at 0 to 2 hours and 2 to
8 hours. Additionally, the result may also support the point that the traditional SampEn can
be a measure of the physiological fatigue during long-term Web browsing.
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
486 HUNG AND JIANG

TABLE 1. Ratios of Highest and Lowest SampEn Calculated for all Subjects

Subject No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

Ratio 1.372 1.513 1.351 1.369 1.663 1.212 1.313 1.322


Subject No. 9 No. 10 No. 11 No. 12 No. 13 No. 14 No. 15 Mean ± S.D.

Ratio 1.434 1.526 1.245 1.518 1.245 1.338 1.391 1.388 ± 0.124

Note. Mean value and standard deviation are in bold.

Unlike the physiological findings, the statistical result of the mean psychological fatigue
score shows that the mean score monotonically increases with the browsing time beyond
4 hours. The correlation between psychological fatigue and physiological fatigue has been
analyzed. At browsing time = 2 and 4 hours, psychological fatigue was weakly relevant to
physiological fatigue (r = 0.121 and 0.055, respectively). At browsing time = 6 and 8 hours,
psychological fatigue and physiological fatigue became increasingly negatively correlated
(r = −0.411 and −0.479, respectively). It seems that when the subjects were getting tired
late in the 8-hour browsing period, one who had lower SampEn reported a higher extent
of subjective fatigue. However, psychological fatigue and physiological fatigue are not
significantly correlated (p > 0.05).
In general, the result in Figure 3 indicates that long-term Web browsing may increase the
entropy (complexity) of the biologic system. To explore the maximum range of SampEn
during the 8-hour Web-browsing task, we normalized the values by calculating the ratio of
the highest SampEn to the lowest for each subject. The 15 subjects were numbered, and the
results are listed in Table 1, where the mean value and the SD for all the subjects appear in
boldface. The ratio was as high as 1.663 for subject No. 5, and as low as 1.212 for subject
No. 6. On average, the browsing task raised SampEn by approximately 38.8%.

3.4. The MSE Result


Costa et al. (2005) used the MSE method to investigate SampEn at various scales from the
24-hour cardiac interbeat interval time series of healthy young subjects, including waking
(free-running) and sleeping periods. SampEn derived from the ECG data of waking periods
increases at small scales and then stabilizes to a relative constant value of 1.45 when the
scale factor increases up to 20. However, during sleeping periods, SampEn monotonically
decreases at large scales. The SampEn values for the waking period are higher than for
the sleeping period. This shows that, under free-running conditions, the cardiac dynamics
of healthy young subjects are the most complex. The decline in SampEn during sleeping
periods is thought to stem from the reduction in long-range correlations. The mean SampEn
values calculated from 8-hour ECG signals at Scales 1 to 20 are plotted in Figure 4. For
all the scales considered, SampEn is lowest (1.37) at scale 1. SampEn increases from Scale
1 to Scale 4, reaches its highest value of 1.79, and then it gradually decreases to 1.52
as the scale factor increases from 5 to 20. Therefore, the trend in SampEn toward scales
used with the 8-hour Web-browsing task is similar to that for healthy young subjects under
free-running conditions at small scales and that for these subjects during sleeping periods at
large scales. It should be noted that the browsing tasks in our experiment are presumed to be
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 487

Figure 4 SampEn against scale factors during the 8 hr browsing task (solid circle = mean, vertical
bar = standard error).

less complex than the tasks performed under free-running conditions. Murata et al. (2001)
stated that symptoms of visual fatigue contribute to general fatigue. They also found that,
during VDT tasks, the pupil diameter decreases with time, which indicates the dominance
of the parasympathetic nervous system and the deceleration in the activity of the autonomic
nervous system (ANS). Thus, the sympathetic–parasympathetic balance of the ANS may
have an influence on cardiac dynamics during the browsing task.
We found that two subjects had dozed off during the browsing task; the result is shown
in Figure 5(a) and 5(b). SampEn did not decrease at large scales and therefore has the same
tendency as in the case of healthy young subjects during waking periods presented by Costa
et al. (2005). In addition, it is notable that, at large scales, the entropy values of the subjects
who had dozed off were higher than those of the other subjects.
The MSE results of the 8-hour browsing task for the groups of undergraduates and
graduates are compared in Figure 6. The trends in SampEn toward scales for undergraduates
and graduates are similar to that for all the subjects, except for the fact that the highest
SampEn for the undergraduates occurs at Scale 5 and not Scale 4. As is illustrated in
Figure 6, the SampEn of the undergraduates is higher than that of graduates. The two
curves are indistinguishable at small scales; they nearly overlap at scales between 2 and
4. Furthermore, the two curves gradually diverge as the scale increases; the difference is
approximately 0.175 at Scale 20. The results in Figure 6 indicate that the entropy values of
undergraduates and graduates are distinguishable at large scales (Student’s t test p < 0.05).
The MSE results for males and females during the 8-hour browsing tasks are plotted in
Figure 7. The trends in SampEn toward scales for males and females are similar to that for all
the subjects, except for the fact that the highest value of entropy for males occurs at Scale 3
and not Scale 4. As is shown in Figure 7, the entropy value of males is higher than that of
females (Student’s t test p < 0.05). Unlike the findings in Figure 6, the result indicates that
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
488 HUNG AND JIANG

Figure 5 (a) SampEn against scale factors for the first subject who dozed off during the 8 hr
browsing task; (b) SampEn against scale factors for the second subject who dozed off during the 8 hr
browsing task.

the entropy values of males and females are most distinguishable at scales between 2 and
4. Furthermore, the distances between the two curves are nearly the same at scales greater
than 5. As for the analysis of the psychological fatigue scores between different groups,
there are no significant differences between males/undergraduates and females/graduates
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 489

Figure 6 SampEn against scale factors for undergraduates (solid circle = mean) and graduates
(solid triangle down = mean) during the 8 hr browsing task (vertical bar = standard error).

Figure 7 SampEn against scale factors for males (solid circle = mean) and females (solid triangle
down = mean) during the 8hr browsing task (vertical bar = standard error).

(p > 0.05). Therefore, the MSE method is more powerful than psychological fatigue scores
in distinguishing between groups.
Figures 6 and 7 show that the cardiac dynamic systems of undergraduates and males are
more complex than those of graduates and females, respectively. Because graduates have
more experience in browsing tasks than do undergraduates, the reason for the SampEn of
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
490 HUNG AND JIANG

Figure 8 SampEn against scale factors for 0–4 hr (solid circle = mean) and 4–8 hr (solid triangle
down = mean) browsing periods (vertical bar = standard error).

undergraduates being larger at large scales may arise from their higher curiosity for the
content on the WWW. Costa et al. (2005) also compared the entropy values of healthy
young subjects with those of elderly subjects during the sleeping period. Their results imply
that the parasympathetic branch of the ANS contributes to the entropy value at small scales.
The fact that the entropy values of males are higher than those of females, especially at
Scales 2 to 4, may be interpreted as being the result of the diverse responses of the ANS for
males and females during browsing tasks.
The MSE results from ECG data for the first half of the browsing period (0 to 4 hours)
and those for the second one (4 to 8 hours) are plotted in Figure 8. At Scale 1, which is the
case of the traditional measure of entropy, the mean SampEn of the second period is larger
than that of the first. This finding indicates that the biologic system has higher complexity
during the second period. However, comparing the results of SampEn at various scales, the
mean SampEn during the second period is slightly higher than that during the first period
only at scales less than 3, and the two curves nearly coincide at most of the other scales.
Furthermore, Student’s t test showed no significant difference between the two curves
(p > 0.05).
From the results and discussions above, the traditional SampEn may have the potential
for investigating physiological fatigue during long-term Web-browsing tasks.

4. CONCLUSION
In this article, we used a simple and noninvasive instrument (Philips IntelliVue MP60 patient
monitor) to gather 8-hour continuous ECG time series data for subjects engaged in long-term
Web-browsing tasks. The HR decreases as the browsing task begins, but slightly increases
after 4.5 hours. This increase in HR may originate from the modulation of the subjects’
discomfort. The psychological fatigue score gradually increases with browsing time, which
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 491

implies that the subjects suffer from more severe subjective fatigue as the browsing task
proceeds. Unlike the trend of the psychological fatigue score, the traditional SampEn
oscillates after 2.5 hours, although it increases at the beginning. Hence, the browsing time
of 2.5 hours may be a clue of physiological fatigue. The oscillation might stem from self-
adjustable mechanisms of the biologic system, and it might somewhat reflect the extent of
physiological fatigue. The statistical analysis indicates that SampEn is significantly different
during 0 to 2 hours and 2 to 8 hours, which means that the states of cardiac dynamics are
different during the two time periods.
The MSE results show that the cardiac dynamics of subjects during Web browsing is
less complex than that of healthy young subjects under free-running conditions. For the
groups of undergraduates and graduates, the SampEn values are discernible at large scales.
The SampEn for males and females are especially distinguishable at scales between 2 and
4. Moreover, the cardiac dynamic systems of undergraduates and males are more complex
than those of graduates and females, respectively.
There are three advantages of SampEn or MSE over the psychological fatigue scores:
(1) The measure of SampEn or MSE is objective whereas the psychological fatigue score
is subjective. (2) SampEn indicates the self-adjustable mechanisms of subjects whereas
the psychological fatigue score monotonically increases during the experiment. (3) The
MSE method is more powerful than psychological fatigue scores in distinguishing between
groups.
In summary, the findings indicate that the traditional SampEn may have the potential for
exploring physiological fatigue during long-term Web-browsing tasks.

ACKNOWLEDGMENTS
We appreciate C.-K. Peng and J.-S. Shieh for valuable discussions and kindly assistance.
We also thank the subjects, 15 Yuan Ze university students, to join our experiment. The
Phillips IntelliVue MP60 patient monitor was provided by J.-S. Shieh.

REFERENCES
Abásolo, D., Hornero, R., Espino, P., Álvarez, D., & Poza, J. (2006). Entropy analysis of the EEG
background activity in Alzheimer’s disease patients. Physiological Measurement, 27, 241–253.
Acharya, U. R., Joseph, K. P., Kannathal, M., Lim, C. M., & Suri, J. S. (2006). Heart rate variability:
A review. Medical & Biological Engineering & Computing, 44, 1031–1051.
Akay, M. (2000). Nonlinear biomedical signal processing, Vol. II, dynamic analysis and modeling.
New York: IEEE Press Series on Biomedical Engineering.
Balci, R., & Aghazadeh, F. (2004). Effects of exercise breaks on performance, muscular load, and
perceived discomfort in data entry and cognitive tasks. Computer and Industrial Engineering, 46,
399–411.
Balocchi, R., Carpeggiani, C., Fronzoni, L., Peng, C. K., Michelassi, C., Mietus, J., & Goldberger,
A. L. (1999). Short and long-term heart rate dynamics in atrial fibrillation. In M. Rienzo,
G. Mancia, G. Parati, A. Pedotti, and A. Zanchetti (Eds.). Methodology and clinical applications
of blood pressure and heart rate analysis. Amsterdam: Ios Press Inc., 91–96.
Beasley, R., Raymond, N., Hill, S., Nowitz, M., & Hughes, R. (2003). eThrombosis: The 21st century
variant of venous thromboembolism associated with immobility. European Respiratory Journal,
21, 374–376.
Bergqvist, U., & Knave, B. G. (1994). Eye discomfort and work with visual display terminals.
Scandinavian Journal of Work and Environmental Health, 20, 27–33.
Boucsein, W., & Thum, M. (1997). Design of work/rest schedules for computer work based
on psychophysiological recovery measures. International Journal of Industrial Ergonomics, 20,
51–57.

Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm


492 HUNG AND JIANG

Chi, C. F., & Lin, F. T. (1998). A comparison of seven visual fatigue assessment techniques using
three data-acquisition VDT tasks. Human Factors, 40(4), 577–590.
Comani, S., Srinivasan, V., Alleva, G., & Romani, G. L. (2007). Entropy-based automated classification
of independent components separated from fMCG. Physics in Medicine and Biology, 52, N87–
N97.
Costa, M., Goldberger, A. L., & Peng, C. K. (2002). Multiscale entropy analysis of complex physio-
logic time series. Physical Review Letters, 89(6), 068102(1–4).
Costa, M., Goldberger, A. L., & Peng, C. K. (2005). Multiscale entropy analysis of biological signals.
Physical Review E, 71, 021906(1–18).
Costa, M., Peng, C. K., Goldberger, A. L., & Hausdorff, J. M. (2003). Multiscale entropy analysis of
human gait dynamics. Physical A, 330, 53–60.
Dillon, A., Kleinman, L., Choi, G. O., & Bias, R. (2005). Visual search and reading tasks us-
ing ClearType and regular screen displays: Two experiments. Accessed November 2005, from:
http://www.ischool.utexas.edu/∼ct/
Dillon, T. W., & Emurian, H. H. (1995). Reports of visual fatigue resulting from use of a video display
unit. Computers in Human Behavior, 11(1), 77–84.
Du, N., Bai, S., Oguri, K., Kato, Y., Matsumoto, I., Kawase, H., & Matsuoka, T. (2005). Heart rate
recovery after exercise and neural regulation of heart rate variability in 30-40 year old female
marathon runners. Journal of Sports Science and Medicine, 4, 9–17.
Duffy, V. G., & Chan, A. H. S. (2002). Effects of virtual lighting on visual performance and eye
fatigue. Human Factors and Ergonomics in Manufacturing, 12(2), 193–209.
Gour, J., Edwards, R., Lemieux, S., Ghassemi, M., Jog, M., & Duval, C. (2007). Movement patterns
of peak-dose levodopa-induced dyskinesias in patients with Parkinson’s disease. Brain Research
Bulletin, 74, 66–74.
Hayano, J., Yamasaki, F., Sakata, S., Okada, A., Mukai, S., & Fujinami, T. (1997). Spectral charac-
teristics of ventricular response to atrial fibrillation. American Journal of Physiology, Heart and
Circulatory Physiology, 273(6), H2811–H2816.
Higuchi, S., Motohashi, Y., Liu, Y., Ahara, M., & Kaneko, Y. (2003). Effects of VDT tasks with a
bright display at night on melatonin, core temperature, heart rate, and sleepiness.Journal of Applied
Physiology, 94, 1773–1776.
Kim, W. S., Yoon, Y. Z., Bae, J. H., & Soh, K. S. (2005). Nonlinear characteristics of heart rate
time series: Influence of three recumbent positions in patients with mild or severe coronary artery
disease. Physiological Measurement, 26, 517–529.
Lewis, M. J., & Short, A. L. (2007). Sample entropy of electrocardiographic RR and QT time-series
data during rest and exercise. Physiological Measurement, 28, 731–744.
Li, G., & Buckle, P. (1999). Current techniques for assessing physical exposure to work-related
musculoskeletal risks, with emphasis on posture-based methods. Ergonomics, 42(5), 674–695.
Mandell, A. J., & Shlesinger, M. F. (1990). Lost choices: Parallelism and topological entropy decre-
ments in neurobiological aging. In S. Krasner (Ed.). The ubiquity of chaos. American Association
for the Advancement of Science Publications, Washington, D.C., 35–46.
Mao, I. F., Huang, J. W., Chen, C. J., Tsay, M. H., Cheng, S. F., Chow, C. K., & Chen, M. L. (1998).
Subjective fatigue in visual display terminal workers. Journal of Occupational Safety and Health,
6(2), 71–85.
Murata, A., Uetake, A., Matsumoto, S., & Takasawa, Y. (2003). Evaluation of shoulder muscular
fatigue induced during VDT tasks. International Journal of Human-Computer Interaction, 15(3),
407–417.
Murata, A., Uetake, A., Otsuka, M., & Takasawa, Y. (2001). Proposal of an index to evaluate visual
fatigue induced during visual display terminal tasks. International Journal of Human-Computer
Interaction, 13(3), 305–321.
Park, M. Y., Kim, J. Y., & Shin, J. H. (2000). Ergonomic design and evaluation of a new VDT
workstation chair with keyboard-mouse support. International Journal of Industrial Ergonomics,
26, 537–548.
Paulus, M. P., Geyer, M. A., Gold, L. H., & Mandell, A. J. (1990). Application of entropy derived from
the ergodic theory of dynamical systems to rat locomotor behavior. Proceedings of the National
Academy of Sciences USA, 87, 723–727.
Peng, C. K., Havlin, S., Hausdorf, J. M., Mietus, J. E., Stanley, H. E., & Goldberger, A. L. (1995).
Fractal mechanisms and heart rate dynamics: Long-range correlations and their breakdown with
disease. Journal of Electrocardiology, 28(1), 59–64.
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
MSE APPROACH TO PHYSIOLOGICAL FATIGUE DURING LONG-TERM WEB BROWSING 493

Pincus, S. M. (1991). Approximate entropy as a measure of system complexity. Proceedings of the


National Academy of Sciences USA, 88, 2297–2301.
Platisa, M. M., & Gal, V. (2006). Dependence of heart rate variability on heart period in disease and
aging. Physiological Measurement, 27, 989–998.
Platisa, M. M., Mazic, S., Nestorovic, Z., & Gal, V. (2008). Complexity of heartbeat interval series in
young healthy trained and untrained men. Physiological Measurement, 29, 439–450.
Richman, J. S., & Moorman, J. R. (2000). Physiological time-series analysis using approximate
entropy and sample entropy. American Journal of Physiology-Heart Circulatory Physiology, 278,
2039–2049.
Seghers, J., Jochen, A., & Spaepen, A. (2003). Posture, muscle activity and muscle fatigue in prolonged
VDT work at different screen height settings. Ergonomics, 46(7), 714–730.
Sheedy, J. E. (1992). VDTs and vision complaints. Information Display, 4(5), 20–23.
Shieh, K. K., Chen, M. H., & Wang, Y. W. (2005). Effects of display medium and luminance contrast
on memory performance and EEG response. International Journal of Industrial Ergonomics, 35,
797–805.
Stüdeli, T., & Menozzi, M. (2003). Effect of subjective and objective workload on asthenopia at VDU
workplaces. International Journal of Occupational Safety and Ergonomics, 9(4), 441–451.
Taiwan Network Information Center. 2007. Internet broadband usage in Taiwan. A Summary Report
of the Jan Survey of 2007, from: http://www.twnic.net.tw/download/200307/96305b.pdf
Takahashi, K., Sasaki, H., Saito, T., Hosokawa, T., Kurasaki, M., & Saito, K. (2001). Combined effects
of working environmental conditions in VDT work. Ergonomics, 44(5), 562–570.
Vuksanovié, V., & Gal, V. (2007). Heart rate variability in mental stress aloud. Medical Engineering
and Physics, 29, 344–349.
Wolf, A., Swift, J. B., Swinney, H. L., & Vastano, J. A. (1985). Determining Lyapunov exponents
from a time series. Physica, 16(D), 285–317.
Zeng, W., & Glass, L. (1996). Statistical properties of heartbeat intervals during atrial fibrillation.
Physical Review E, 54, 1779–1784.

Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm

Vous aimerez peut-être aussi