Information Technology Reference
In-Depth Information
be useful to enhance the reliability of system decisions and to estimate the dy-
namic evolution of physiological age relative to the chronological age.
The structure of the paper is as follows. Section 2 describes the experimental
setup. Complexity measures are explained in section 3. Our results are presented
and discussed in section 4. Finally, we summarize our conclusions in section 5.
2 Experimental Setup
2.1 Data Acquisition and Preprocessing
A sample of 126 healthy subjects (ages 9-75 yr) was initially selected. The volun-
teers were encouraged to maintain their usual lifestyles. A Thermochron iButtom
miniaturized data logger (Dallas, Maxim) was placed on the wrist of the non-
dominant hand with the help of a wristband. Temperature was logged every 10
minutes for a variable number of days. To ease the interpretation of results we
will uses days (
d
) as time unit and cycles/day (
c/d
) as frequency unit. Thus, the
sampling period is
T
=0
.
0069
d
and the sampling frequency is
F
s
= 144
c/d
.
The signal was filtered with a 8th order Chebyshev Type I low-pass antialiasing
filter with a cut-off frequency of 57
.
60
c/d
(0
.
8
F
Nyquist
).
The recorded signal exhibits occasional artifacts displayed as large tempe-
rature drops, that appear when the subject removes the wristband or when
the sensor is badly positioned for some time. A simple preprocessing algorithm
removes large artifacts and replaces missing data with interpolate data. A reli-
ability index was defined as the ratio between the number of replaced samples
and total number of samples. Sixty records with reliability index smaller than
90% were discarded. Moreover, 13 short records (duration less than 7
d
)were
also discarded. Larger records were clipped to the seven central days. Thus, the
final sample consists of 53 records with a duration of 7
d
(1008 samples).
×
2.2 Statistical Analysis and Surrogates
Most of the spectral energy of recorded signals is confined below 5
.
46
1
.
83
c/d
.
This was determined by applying a 32 samples median filtered periodogram.
Non filtered periodogram shows a large peak frequency of 1
c/d
for most of the
records, corresponding to a normal entrained circadian rhythm. Previous studies
reveal the existence of weak nonstationarities on circadian rhythms [12]. We
conducted a study based on the application of 1st order symlets, an orthogonal
discrete wavelet transform (DWT), focused on the ranges 0-1.13 and 1.13-9.02
c/d
. Most of the energy (99
.
95
±
0
.
05%) is concentrated in the first range.
A 4th order Chebyshev Type I low-pass filter with a cut-off frequency of
5
.
46
c/d
was applied to the raw signal. A different filtering approach was also
tested. We applied the 5th order Daubechies DWT and reconstructed the A4 ap-
proximation. This resolution level roughly correspond to a LP cut-off frequency
of 6.04
c/d
. We will denote the raw signal by
s
, the first filtered signal by
sflp
and the wavelet reconstruction by
sfa4
.
±
Search WWH ::
Custom Search