Biomed. Eng.-Biomed. Tech.
2017; aop
Jayanthi Thiruvengadam* and Anburajan Mariamichael
A preliminary study for the assessment
of hypertension using static and dynamic
IR thermograms
DOI 10.1515/bmt-2016-0237
Received July 4, 2016; accepted March 1, 2017
Abstract: Structural changes in blood vessels occur due
to prolonged hypertension. Early detection of blood pres-
sure (mm Hg) is essential for disease prevention. The aim
of this work is to propose a computer-aided diagnostic
(CADx) model for the diagnosis of hypertension using var-
iables derived from non-contact static and dynamic ther-
mal imaging in comparison with the pulse wave velocity
(PWV)-derived parameters. Static and dynamic infrared
(IR) thermograms of selected skin areas of the body from
known hypertensive (n = 14) and age- and sex-matched
normal subjects were captured. The average skin surface
temperature [SST (°C)] of selected skin areas of the body
was calculated from a static IR thermogram. After denois-
ing the dynamic IR thermogram using wavelets, the sta-
tistical variables power, mean, standard deviation (SD),
variance, skewness and kurtosis were calculated. The var-
iables derived from both static and dynamic thermograms
were used to develop the CADx model. The performance of
the CAD model was also tested by feature selection using
principal component analysis (PCA). An accuracy of 75%
(sensitivity = 78.6%, specificity = 71.4%) could be achieved
with the average SST (°C) of the static IR thermogram
alone. The statistical variables derived from the dynamic
IR thermogram alone gave an accuracy of 82% (and 85%
after feature selection by PCA), whereas the accuracy
using standard methods like variables derived from PWV
was only 71.4% (with and without feature selection). The
highest accuracy of 89% could be achieved by combining
variables like average SST (°C) measured from static and
dynamic IR thermograms and PWV-derived variables.
*Corresponding author: Dr. Jayanthi Thiruvengadam, Assistant
Professor, Department of Biomedical Engineering, SRM University,
SRM Nagar, Kattankulathur, Kancheepuram District, Chennai
603203, Tamil Nadu, India, Phone: +91 98404 90315,
E-mail: jayanthi_20@yahoo.com; jayathi.t@ktr.srmuniv.ac.in
Anburajan Mariamichael: Department of Biomedical Engineering,
SRM University, SRM Nagar, Kattankulathur, Kancheepuram District,
Chennai 603203, Tamil Nadu, India
Keywords: hypertension; pulse wave velocity; thermal
imaging; vascular dysfunction; wavelet denoising.
Introduction
Hypertension is the most common modifiable risk factor
for cardiovascular disease (CVD). The world health
organization (WHO) estimates that in 2025, there will be
1.56 billion adults living with hypertension [19]. World
Health Statistics 2012 reports that one in three adults
has hypertension. In India, hypertension is the direct
cause of 57% of all stroke deaths and 24% of all coronary
heart disease [9]. Arterial vessel wall stiffness, which is a
measure of elasticity of the vessel walls, is an important
determinant of CVD [14]. It is a change in pressure for
unit change in volume, which is the important determi-
nant of hypertension. The Young’s modulus of the vessel
wall increases with pressure experienced by the arter-
ies. It is the cause of CVD that is commonly quantified
by PWV.
Radiation emitted by the object depends on its
temperature and emissivity. The human skin surface is
almost equivalent to the surface of a black body, with
an emissivity of 0.98 regardless of race and origin. Thus,
the infrared (IR) spectrum can be used to quantify skin
surface temperature [SST (°C)] with good sensitivity. IR is
gaining importance for mapping blood vessel. The vas-
culature underneath the skin is responsible for the SST
(°C) of the body [15]. The pulsatile flow of blood causes a
dominant variation in the SST (°C). Hence, IR radiation
can be used for mapping the vasculature underneath.
The IR thermogram of the blood vessel is, thus, used as
a template for an individual for biometric applications
[4]. The SST (°C) variation or waveform was found to be
correlated with the pulsatile flow of blood [7]. The tem-
perature waveform was filtered with a frequency lower
than 0.67  Hz (40 bpm) and higher than 1.67  Hz (100
bpm). The temperature signal and the pulse waveform
were found to be similar after filtering. The temperature
signal was also found to be the strongest in the areas
where superficial arteries could be located. Hence, the
Bereitgestellt von | De Gruyter / TCS
Angemeldet
Heruntergeladen am | 18.07.17 15:56
2      J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms
forehead and the lateral neck areas, where the carotid
artery and its branches are situated, are the common
areas for extracting the temperature waveform. In an
antithrombotic therapy study, vasodilators reduced the
core temperature. Kenney et  al. pointed out the pos-
sibility of reduced mean arterial blood with increased
blood flow to the skin due to vasodilators [3, 11]. In order
to frame an efficient model, the signal parameters are
derived from the discrete SST (°C) waveform. The charac-
teristics of the random signal can be described by statis-
tical parameters such as energy, power, mean, standard
deviation (SD), variance, skewness and kurtosis. Energy,
mean and power depict how the signal amplitude varies.
The probability distribution of the signal and the varia-
tion in frequency is described by SD, variance, skewness
and kurtosis [1].
Accurate and reproducible measurement of hyper-
tension is a critical need for the assessment of high
blood pressure. Blood pressure is monitored using an
oscillometric sphygmomanometer and a stethoscope. It
determines the mean arterial pressure in the upper bra-
chial artery. The measurement has to be repeated for the
confirmation of the disease, which is the only available
method clinically [8, 12]. Hence, the aim of this study
was to investigate the significance of the measured
variables, namely (i) average SST (°C) from the static IR
thermogram, (ii) average SST (°C/min) measured from
the dynamic IR thermogram, (iii) statistical variables
extracted from the dynamic IR thermogram and (iv)
mechanical blood flow from PeriScope (Genesis Medical
Systems, Hyderabad, India) in the evaluation of hyper-
tension. A CADx model with high accuracy was also
developed using different sets of the above-­mentioned
measured variables by testing and comparing with
feature reduction using principal component analy-
sis (PCA) and a feedforward backpropagation neural
network classifier.
Methods
Measurements
A free screening camp for hypertension was organized during Sep-
tember 2013. The total number of registered subjects of both sexes
was 300, and their age ranged from 18 to 70 years. An informed con-
sent form was obtained from each subject. A detailed questionnaire
(Appendix I) prepared for this study was administered to each sub-
ject and the details were noted. Of the total subjects enrolled, 40 were
known cases of fever, neurological disorders, diabetes mellitus, thy-
roid abnormalities and nephropathy and, thus, were excluded from
this study. Of the remaining, 14 hypertensive and 14 normal age- and
sex-matched subjects were chosen for the study. The study group is
given as follows:
Total number of subjects in the study population: 28
–– Group I: Known hypertensive [n = 14 (9  men/5  women),
mean ± SD age = 39.3 ± 10.5  years), average number of years
since hypertension = 2 ± 0.6 years];
–– Group II: Normal age- and sex- matched subjects [n = 14
(9 men/5 women)], mean ± SD age = 39.7 ± 12.2 years];
In each subject, mechanical blood flow variables were measured
using a PC-based PeriScope device. It is an oscillometry-based blood
pressure monitoring with PC-based acquisition. The following vari-
ables were calculated: (i) carotid-femoral pulse wave velocity (cf-
PWV), (ii) brachial PWV (ba-PWV) – bilateral, (iii) Arterial stiffness
index (ASI) – bilateral, (iv) ankle ASI – bilateral, (v) ankle-brachial
index (ABI) and (vi) augmentation index (AIx). PWV is calculated by
measuring the time taken for the pressure waves to travel between
two points where the recording is done. ASI is the measure of loss of
elasticity of the vessel wall. AIx is used to assess the reflected wave
pattern, calculated as AIx% =[(ΔP/PP) × 100], where ΔP is the differ-
ence in peak systolic blood pressure (SBP) and the early upstroke of
the reflected wave and PP is the pulse pressure.
Moreover, both static and dynamic IR thermograms of differ-
ent body regions were captured in each subject using a standard IR
thermal camera (FLIR A305sc, FLIR Systems, Inc., Wilsonville, OR,
USA; thermal sensitivity >0.05°C at 30°C, with accuracy ±2°C). A
systematic protocol as proposed by the International Association of
Certified Thermographers was maintained for capturing the thermo-
gram. The clinical trial was conducted in special metabolic clinical
trial ward constructed under the Good Clinical Practice regulation,
where the room temperature was maintained at 20°C. The relative
humidity was 50% and emissivity was set at 0.98. The camera cali-
bration program uses temperature references that are calibrated
annually and traceable to the National Institute of Standards of
Testing. It also has an option to perform internal calibration when
needed. All the subjects were made to sit for 15 min in the temper-
ature-controlled waiting room, which reduces the effect of an envi-
ronment temperature of the skin. The thermal camera was placed on
a tripod at a distance of 1 m from the subject, who was asked to sit
on a fixed chair to facilitate the IR thermogram recording process.
The subjects were uncovered below the knee, with no metal orna-
ments of any kind on the hands or face. The camera was focused
for 1 min on the following selected skin areas of the body and static
IR thermograms were obtained: (i) forehead [anterior-posterior
(AP) view, (ii) neck (lateral view of bilateral sides), (iii) forearm (AP
view of bilateral sides) and (iv) foot [both AP and posterior-anterior
(PA) views of bilateral sides]. In the aforementioned body regions,
dynamic IR thermograms were recorded for 1  min at the rate of 9
frames/s and were stored in a.fcf format, whereas the static IR ther-
mograms were stored in a .jpg format.
IR thermogram analysis
Static IR thermogram: In each acquired IR thermogram of the
selected body regions, a suitable region of interest (ROI) was selected
using the “rectangle” select tool, available in the software FLIR
ReserachIR (FLIR Systems, Inc., Wilsonville, OR, USA), and an aver-
age SST (°C) of the corresponding region was measured [6].
Bereitgestellt von | De Gruyter / TCS
Angemeldet
Heruntergeladen am | 18.07.17 15:56
 J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms      3

Dynamic IR thermogram: In each video file of the parts of the
selected body region, a suitable ROI was chosen using the “rectan-
gle” select tool, available in the software FLIR ReserachIR. In the
selected ROI, the average measured SST (°C) for 1 min was imported
into a.csv file format. Using this, the average SST (°C/min) of the
ROI was calculated from each corresponding region. Further, each
IR dynamic signal was decomposed using the discrete wavelet
transforms (DWTs) Daubechies wavelet 8 (db8) and Haar wavelet.
The estimated mean values of the signal-to-noise ratio (SNR) of
the original video thermal signal of the selected body regions were
found to be 10.6  dB (±3.09), whereas the calculated mean values
were found to be 50.2 (±4.5 dB) and 14.03 dB (±4.3) after denoising
using db8 – level 5 and the Haar wavelet, respectively. It was found
that the calculated mean values of SNR were greater in the video
thermal signal after denoising by db8  when compared to the sig-
nal using the Haar wavelet (Table 1). Figure 1 shows the denoising
steps in the MATLAB wave menu. After denoising, the signal was
stored and the following statistical variables were derived, namely
(i) energy, (ii) power, (iii) mean, (iv) variance, (v) SD, (vi) kurtosis
and (vii) skewness.

Table 1: Comparison of SNR (dB) of the dynamic IR thermogram of selected body regions after denoising by two different wavelet denoising
(Haar and Daubechies) methods.

S. No   Thermal signal of selected skin areas   SNR (dB)

of the body acquired from normal  
Obtained   Denoised by   Denoised by Haar
subjects (dynamic IR thermogram)
original signal DWT-db8 at level 5 wavelet at level 5

1   Forehead   12.28   47.96   17.99

2   Neck      
   Left   13.28   57.28   19.40
   Right   10.1   46.63   15.3
3   Forearm      
   Left   11.488   52.07   16.6
   Right   15.52   54.88   15.4
4   Foot anterior      
   Left   11.07   52.22   15.7
   Right   9.2   51.79   10.5
5   Foot posterior      
   Left   6.18   44.26   7.8
   Right   6.18   44.56   7.6

Mean ± SD   10.6 ± 3.09   50.2 ± 4.5   14.03 ± 4.3

Figure 1: MATLAB Wavemenu screenshot showing the steps in denoising using a wavelet (DWT-db8) dynamic IR thermogram signal of the
forehead in a normal subject.

Bereitgestellt von | De Gruyter / TCS

Angemeldet
Heruntergeladen am | 18.07.17 15:56
4      J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms

CADx model for hypertension individually using a backpropagation neural (BPN) network classi-
fier. The same was repeated with feature selection by PCA (SPSS Ver-
sion 17.0). The results were tabulated and compared.
In each subject, the total number of variables measured was 94
and its break-up is summarized as follows: (i) 9 [average SST (°C)
of selected skin areas of the body using the static IR thermogram],
(ii) 9 [average SST (°C/min) of selected skin areas of the body using
the dynamic IR thermogram], (iii) 63 [7 statistical variables extracted Results
from the thermal signal of the selected skin areas of the body using
the dynamic IR thermogram (9 regions × 7 = 63 variables)] and (iv)
13 (mechanical variables of blood flow using PeriScope). Using the Statistical comparison of known
above variables, the CADx model was developed following two dif- ­hypertensive and normal subject
ferent approaches.
(a) Individual approaches with and without feature selection by
Table 2 shows the statistical comparison of demographic
PCA and
(b) Combinational approaches with and without PCA. and PeriScope variables for known hypertensive and
normal subjects. Except for ASI (mm Hg) and ABI, all
The performance of each approach with a particular set of variables other parameters showed differences between the normal
in the evaluation of hypertension with good accuracy was tested and hypertensive groups. Figure 2 shows the dynamic

Table 2: Comparison of demographic and mechanical variables of blood flow between known hypertensive and normal subjects.

S. No  Variables       Group I: Known  Group II: Age- and sex-  Statistical
    hypertensive [n = 14  matched normal [n = 14 significance
(9 men/5 women), (9 men/5 women), (p-Value)
mean ± SD mean ± SD
age = 39.3 ± 10.5 years] age = 39.7 ± 12.2 years]

Mean ± SD Mean ± SD

I.   Demographic details
  Body weight (kg)       74.07 ± 12.72  67.43 ± 11.82  NS
  Body height (cm)       166.29 ± 8.1  161.07 ± 7.88  NS
II.   Mechanical variables of blood flow (PeriScope)
  a. Directly measured
    HR (bpm)       77.7 ± 9.09  72.14 ± 8.62  NS
    SBP (mm Hg)       140.29 ± 10.05  118.64 ± 9.54  <0.01
    DBP (mm Hg)       81.71 ± 13.94  68.14 ± 5.81  <0.05
    Pulse pressure (mm Hg)       58.5714 ± 9.24  50.78 ± 8.94  <0.05
  b. Indirectly derived
    Central pressure    Aortic SBP (mm Hg)   117.69 ± 14.93  99.85 ± 9.11  <0.01
     Aortic pulse pressure (mm Hg)  37.42 ± 7.88  27.92 ± 7.11  <0.01
     Aortic DBP (mm Hg)   81.35 ± 11.3  68.35 ± 4.79  <0.05
     Aortic augmentation pressure   7 ± 4.54  2.57 ± 4.39  <0.05
(mm Hg)
  Arm    Brachial ASI (mm Hg)   26.05 ± 6.26  23.73 ± 5.07  NS
  Leg    Ankle ASI (mm Hg)   35.43 ± 6.32  32.50 ± 4.85  NS
  Neck to leg    cf-PWV (cm/s)   876.9 ± 225.33  658.22 ± 131.24  <0.05
  Hand to leg    ba-PWV (cm/s)   1332.33 ± 270.39  1069.91 ± 157.5  <0.05
  Leg to hand    ABI   1.09 ± 0.16  1.13 ± 0.07  NS
III.   Statistical variables derived from dynamic thermal imaging
  Neck left mean       3.51 ± 0.07  3.49 ± 0.1  <0.05
  Neck right    SD   2.52 ± 0.32  2.75 ± 0.22  <0.05
     Variance   5.04 ± 0.65  5.5 ± 0.45  <0.05
     Kurtosis   1.51 ± 0.81  1.01 ± 0.2  <0.05
  Forearm left mean       3.46 ± 0.08  3.03 ± 0.1  <0.05
  Foot posterior left kurtosis       0.98 ± 0.18  0.96 ± 0.24  <0.05
a
NS, Not significant.

Bereitgestellt von | De Gruyter / TCS

Angemeldet
Heruntergeladen am | 18.07.17 15:56
 J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms      5

IR thermogram of the average SST (°C/min) of the left With feature selection
forearm of a known hypertensive and normal subject in
MATLAB before and after noise removal. The variables obtained from the aforementioned differ-
ent approaches were utilized in the automated evalu-
ation of hypertension. It was found that the following
three methods had the highest accuracy of 89%: (a) static
CADx model for hypertension
IR thermogram and statistical variables with PCA (sen-
sitivity = 92.9% and specificity = 78.6%), (b) dynamic IR
Without feature selection
thermogram and PeriScope variables with PCA (sensitiv-
ity = 85.7% and specificity = 92.9%) and (c) static IR ther-
All the variables obtained from the aforementioned dif-
mogram, dynamic IR thermogram and statistical variables
ferent approaches were utilized in the automated evalua-
without PCA (sensitivity = 85.7% and specificity = 92.9%).
tion of hypertension. It was found that the approach using
On the other hand, the approach with statistical and Peri-
average SST (°C) using the static IR thermogram, average
Scope variables alone had the lowest accuracy of 57%,
SST (°C/min) using the dynamic IR thermogram and sta-
whose sensitivity and specificity were 71% and 60.7%,
tistical variables derived from the dynamic IR thermogram
respectively. The accuracy, sensitivity and the specificity
had the highest accuracy of 89% (sensitivity = 85.7%,
after feature selection for all the above approaches are
specificity = 92.9%). On the other hand, the approach with
listed in Table 3B.
average SST (°C/min) using the dynamic IR thermogram
and mechanical variables of blood flow measured using
PeriScope had the lowest accuracy of 42.9%, whose sensi-
tivity and specificity were 64.3% and 21.4%, respectively. Discussion
The accuracy, sensitivity and specificity of the neural
network tested for different approaches are presented in Elevated heart rate is a common feature in hypertension
Table 3A. [5, 10, 17]. Heart rate increases with peripheral blood

Figure 2: Dynamic IR thermogram of the left forearm and its denoised thermal signal analysis using DWT-db8 in a sample of a normal and a
hypertensive subject.

Bereitgestellt von | De Gruyter / TCS

Angemeldet
Heruntergeladen am | 18.07.17 15:56
 Table 3A: Performance analysis of different CADx models using BPN network in the evaluation of hypertension, including all the measured variables.

Approach  Various approaches for CADx model   No of  Sensitivity %  Specificity %  Accuracy %
variables

(i)   Average SST (°C) using static IR thermogram   9  78.6  71.4  75

(ii)   Average SST (°C/min) using dynamic IR thermogram   9  57.1  71.4  64.3
(iii)   Statistical variables derived from dynamic IR thermogram   63  92.9  71.4  82.1
(iv)   Mechanical variables of blood flow measured using PeriScope   13  75.8  64.3  71.4
(v)   Average SST (°C) using static IR thermogram and average SST (°C/min) using dynamic IR thermogram   18  64.3  78.6  71.4
(vi)   Average SST (°C) using static IR thermogram and statistical variables derived from dynamic IR thermogram   72  92.9  85.7  85.3
(vii)   Average SST (°C) using static IR thermogram and mechanical variables of blood flow measured using PeriScope   22  42.9  85.7  64.3
(viii)   Average SST (°C/min) using dynamic IR thermogram and statistical variables derived from dynamic IR thermogram  72  92.9  78.6  85.7
(ix)   Average SST (°C/min) using dynamic IR thermogram and mechanical variables of blood flow measured using   22  64.3  21.4  42.9
PeriScope
(x)   Statistical variables derived from dynamic IR thermogram and mechanical variables of blood flow measured using   76  57.1  64.3  60.7
PeriScope
(xi)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR thermogram and statistical   81  85.7  92.9  89.3
variables derived from dynamic IR thermogram
(xii)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR thermogram and mechanical   31  21.4  64.3  42.9
variables of blood flow measured using PeriScope
(xiii)   Average SST (°C/min) using dynamic IR thermogram, statistical variables derived from dynamic IR thermogram   85  78.6  42.9  60.7
and mechanical variables of blood flow measured using PeriScope
6      J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms

(xiv)   Average SST (°C) using static IR thermogram, statistical variables derived from dynamic IR thermogram and   85  57.1  71.4  64.3
mechanical variables of blood flow measured using PeriScope
(xv)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR thermogram, statistical   94  78.6  64.3  71.4
variables derived from dynamic IR thermogram and mechanical variables of blood flow measured using PeriScope

Heruntergeladen am | 18.07.17 15:56

Bereitgestellt von | De Gruyter / TCS
Angemeldet
 Table 3B: Performance analysis of different CADx models using BPN in the evaluation of hypertension with feature selection by PCA.

Approach   Different approaches for CADx model   No. of variables   Sensitivity %   Specificity %   Accuracy %

(i)   Average SST (°C) using static IR thermogram   Not available as the number of variables is very less
(ii)   Average SST (°C/min) using dynamic IR thermogram        
(iii)   Statistical variables derived from dynamic IR thermogram   33  78.6  92.9  85.7
(iv)   Mechanical variables of blood flow measured using PeriScope   12  78.6  71.4  71.4
(v)   Average SST (°C) using static IR thermogram and average SST (°C/min) using dynamic IR   13  78.6  85.7  82.1
thermogram
(vi)   Average SST (°C) using static IR thermogram and statistical variables derived from   32  92.9  78.6  89.7
dynamic IR thermogram
(vii)   Average SST (°C) using static IR thermogram and mechanical variables of blood flow   20  64.3  78.6  71.4
measured using PeriScope
(viii)   Average SST (°C/min) using dynamic IR thermogram and statistical variables derived from   62  85.7  78.6  82.1
dynamic IR thermogram
(ix)   Average SST (°C/min) using dynamic IR thermogram and mechanical variables of blood   12  85.7  92.9  89.3
flow measured using PeriScope
(x)   Statistical variables derived from dynamic IR thermogram and mechanical variables of   22  71.4  42.9  57.1
blood flow measured using PeriScope
(xi)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR   19  71.4  85.7  78.6
thermogram and statistical variables derived from dynamic IR thermogram
(xii)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR   8  78.6  50  64.3
thermogram and mechanical variables of blood flow measured using PeriScope
(xiii)   Average SST (°C/min) using dynamic IR thermogram, statistical variables derived from   23  35.7  85.7  60.7
dynamic IR thermogram and mechanical variables of blood flow measured using PeriScope
(xiv)   Average SST (°C) using static IR thermogram, statistical variables derived from dynamic IR   48  78.6  57.1  67.9
thermogram and mechanical variables of blood flow measured using PeriScope
(xv)   Average SST (°C) using static IR thermogram, average SST (°C/min) using dynamic IR   25  78.6  71.4  75
thermogram, statistical variables derived from dynamic IR thermogram and mechanical
variables of blood flow measured using PeriScope

Heruntergeladen am | 18.07.17 15:56

Bereitgestellt von | De Gruyter / TCS
Angemeldet
J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms      7
8      J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms
pressure. The dysregulation of the autonomic function
in hypertension causes an increased heart rate. In our
study, heart rate showed a statistically significant posi-
tive correlation with the average SST (°C/min) of the left
forearm (r = 0.427, p < 0.05) measured using the dynamic
IR thermogram, and the same was significantly higher
by 3.1% in known hypertensive subjects when com-
pared to age-and-sex- matched normal subjects. All the
derived mechanical variables of blood flow were higher
in hypertensive subjects, of which the average ba-PWV
showed the highest percentage increment (26%) when
compared to the normal subjects. The radial artery in
the hand is the main artery that reflects changes in the
central blood pressure. Central pulse pressure was also
found to be highly correlated with carotid intima-media
thickness (IMT) and plaque score [16]. Hence, the radial
artery is better in echoing the central blood pressure
[13]. In our study, the average SST (°C) of the forehead
and the left side of neck measured from the dynamic
IR thermogram significantly correlated with ankle ASI
(r = − 0.413, p < 0.05) and brachial ASI (r = 0.401, p < 0.05),
respectively.
When blood pressure increases, the changes in
blood flow and resistivity to the blood flow are altered
[2]. Hence, a higher value of SD, variance and kurto-
sis indicate drastic variations in the SST (°C) waveform
when compared to normal blood flow. The statistical vari-
able SD derived from the dynamic IR thermogram from
the right side of the neck, was positively correlated with
the systolic BP and aortic systolic BP, and the same was
also significantly higher in known hypertensive than in
normal subjects. The percentage increment was highest
(35%) for kurtosis of the right side of the neck in known
hypertensive subjects when compared to that in normal
subjects. Hence, feature extraction can be further used for
the automated classification of hypertensive subjects. To
the best of our knowledge, the literature review revealed
that no study had been done to compare the statistical
variables derived from the dynamic IR thermogram and
the mechanical variables of blood flow in hypertensive
subjects.
Wang et  al. in their study used 13 variables (BP,
dietary pattern, smoking and blood biochemical markers)
in their CADx model using a neural network to predict
hypertension. They collected those variables from about
308,711 subjects. Using feedforward BPN network with
different hidden layers 8, 9, 10 and 11, the accuracy in
the evaluation of the condition was reported as 72.04%,
72.06%, 71.91% and 72.12%, respectively [18]. In this
study, the accuracy of the developed CADx model using
a BPN network using the following set of all variables
without any other combinational approach were 71.4%,
75%, 64.3% and 82.1%, respectively: (i) mechanical vari-
ables of blood flow using PeriScope (standard method),
(ii) average SST (°C) of selected skin areas of the body
using static IR thermograms, (iii) average SST (°C/min)
of selected skin areas of the body using dynamic IR ther-
mograms and (iv) statistical variables extracted from
the thermal signal of dynamic IR thermograms. By com-
parison, the CADx model with statistical variables of the
thermal signal had the highest accuracy when compared
to the standard method.
When the developed CADx model was implemented
using the following two approaches after variable
selection by PCA, an even better accuracy of 89% was
achieved: (i) average SST (°C) of selected skin areas of
the body using static IR thermograms and statistical vari-
ables extracted from the thermal signal of using dynamic
IR thermograms and (ii) average SST (°C/min) of selected
skin areas of the body using dynamic IR thermograms
and mechanical variables of blood flow using PeriScope.
Of the two approaches, the first approach using IR (both
static and dynamic) thermograms is the preferred CADx
model for the evaluation of hypertension without the
measurement of SBP.
Conclusion
The statistically significant correlation of the mechanical
blood flow variables with the average SST (°C) of selected
body regions such as the left side of the neck, the right and
left forearms and the anterior of the right and left feet indi-
cate the efficacy of dynamic IR thermograms in depicting
the nature of blood flow. The significant difference in
mean values of the statistical variables derived from the
dynamic IR thermogram, such as left mean of the neck,
right SD, kurtosis and variance of the neck, left mean of
the forearm and posterior kurtosis of the foot, is useful in
creating a CADx model for evaluating hypertension. To the
best of our knowledge, no studies have been done so far to
identify hypertensive individuals using variables derived
from static IR thermograms and dynamic IR thermograms
without the inclusion of systolic BP. Thus, the variables
extracted from IR thermal imaging can also be employed
in a CADx model for the evaluation of hypertension with
better accuracy.
Conflict of interest statement: There are no conflicts of
interest.
Funding: This work was not funded by any external
institution.
Bereitgestellt von | De Gruyter / TCS
Angemeldet
Heruntergeladen am | 18.07.17 15:56
 J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms      9
Appendix
Questionaire used in the camp for ­collecting information
about the subjects enrolled in the camp
Questionnaire for Cardiovascular Disease Study
I.D No:
1. Name :
2. Age :
3. Sex : Male ○ Female ○
4. Contact address :
5. Phone Number :
6. Occupation :
7. Family Monthly Income : Rs.
8. Marital status : Married/Unmarried
9. Number of Children :
10. Food intake :
–– Vegetarian/Non-vegetarian
11. Physical activity
(a) REGULAR Exercise + Strenuous work
(b) MODERATE Exercise/Strenuous work
(c) MILD Exercise/Strenuous work
(d) No exercise and sedentary work
12. Known for the following diseases:
–– Known Diabetic: Yes/No
(a) If “Yes”, answer the following:
(b) Type of diabetic: I/ or II
(c) The duration of the disease: ___ years
(d) Are you taking your medications/insulin
every day? Yes/No
13. Previous heart attack/myocardial infarction?
Yes/No
–– If “Yes”, duration of the disease: ___ years
14. Previous stroke? Yes/No
–– If “Yes”, duration of the disease: ___ years
15. Do you have a High Pressure: Yes/No
–– If “Yes”, duration of the disease:
–– Are you taking hypertension drugs: Yes/No
16. Do you have any following complaints:
–– Tuberculosis: Yes/No
–– Arthritis: Yes/No
–– Asthma: Yes/ No
–– Thyroid problems: Yes/No
17. Have you ever had any of the following operations
or procedures related to your heart?
–– Coronary artery bypass surgery (open heart sur-
gery): Yes/No
–– Coronary angioplasty (“balloon” heart proce-
dure): Yes/No
–– Heart catheterization (angiogram): Yes/No
18. Any Surgery underwent Details:
19. Family Positive History:
–– Diabetes mellitus: Yes/No
–– If yes mother/father/both
–– Cardiovascular diseases: Yes/No
–– Diabetes + Cardiovascular disease: Yes/No
–– Stroke: Yes/No
–– Cancer: Yes/No
20. For Female ONLY:
–– Age at menarche (Years):
–– Menstrual cycle: ___ days/month
–– Age at menopause (Year):
–– History of menopause: Natural/Hysterectomy
–– Have you used contraceptive pills: Yes/No
(a) If “Yes”, how many pills were used so for? _____
–– Have you undergone hormone replacement ther-
apy (HRT):
21. Smoking behaviour:
–– Smoker? Yes/No;
(a) If “Yes”, answer the following:
(i) Smoking every day/OR occasional
(ii) Number of cigarettes per day smoked:
(iii) Number of years as a smoker:
(iv) If you stopped smoking, year of smoking
cessation:
Alcohol consumption:
–– Drink alcohol: Yes/No
(a) If “Yes”, specify the type: Beer/Whisky/Rum/
other hot drinks
(i) Volume of liquor: _____ ml
(ii) Frequency: daily/weekly/monthly/occasional
(iii) Number of years as a drinker:
(iv) If you stopped drinking, year of drinking
cessation:
Tobacco:
–– Chewing tobacco: Yes/No
(a) If “Yes”, the frequency of usage: day/week/
month/occasional
(b) Number of years as a tobacco user:
(c) If you stopped chewing tobacco, year of cessation:
Bereitgestellt von | De Gruyter / TCS
Angemeldet
Heruntergeladen am | 18.07.17 15:56
10      J. Thiruvengadam and A. Mariamichael: Evaluation of hypertension using IR thermograms
References
[1] Acharya UR, Vidya KS, Ghista DN, Lim WJ, Molinari F, Sankarana-
rayanan M. Computer-aided diagnosis of diabetic subjects by
heart rate variability signals using discrete wavelet transform
method. Knowl-Based Syst 2015; 81: 56–64.
[2] Bogren HG, Klipstein RH, Mohiaddin RH, et al. Pulmonary artery
distensibility and blood flow patterns: a magnetic resonance
study of normal subjects and of patients with pulmonary arterial
hypertension. Am Heart J 1989; 118: 990–999.
[3] Bruning RS, Dahmus JD, Kenney WL, Holowatz LA. Aspirin and
clopidogrel alter core temperature and skin blood flow during
heat stress. Med Sci Sport Exer 2013; 45: 674.
[4] Buddharaju P, Pavlidis IT, Tsiamyrtzis P, Bazakos M. Physiology-
based face recognition in the thermal infrared spectrum. IEEE
Trans Pattern Anal Mach Intell 2007; 29: 613–626.
[5] Fabian V, Havlik J, Dvorak J, et al. Differences in mean arterial
pressure of young and elderly people measured by oscilometry
during inflation and deflation of the arm cuff. Biomed Eng/
Biomed Tech 2016; 61: 611–621.
[6] FLIR ResearchIR 4 User Guide, Document number: 29354-000,
Rev.3, 2015.
[7] Garbey M, Merla A, Pavlidis I. Estimation of blood flow speed
and vessel location from thermal video. InComputer Vision and
Pattern Recognition, 2004. CVPR 2004. Proceedings of the 2004
IEEE Computer Society Conference on 2004 Jun 27 (Vol. 1, pp.
I–356). IEEE.
[8] Garbey M, Sun N, Merla A, Pavlidis I. Contact-free measurement
of cardiac pulse based on the analysis of thermal imagery. IEEE
Trans Biomed Eng 2007; 54: 1418–1426.
[9] Gupta R. Trends in hypertension epidemiology in India. J Hum
Hypertens 2004; 18: 73–78.
[10] Julius S, Palatini P, Kjeldsen SE, et al. Usefulness of heart rate
to predict cardiac events in treated patients with high-risk
systemic hypertension. Am J Cardiol 2012; 109: 685–692.
[11] Kenney WL, Stanhewicz AE, Bruning RS, Alexander LM. Blood
pressure regulation III: what happens when one system must
serve two masters: temperature and pressure regulation? Eur J
Appl Physiol 2014; 114: 467–479.
[12] Kenning I, Kerandi H, Luehr D, et al. Hypertension diagnosis
and treatment. Blood Press 2014; 140: 90.
[13] Nelson MR, Stepanek J, Cevette M, Covalciuc M, Hurst RT, Tajik
AJ. Noninvasive measurement of central vascular pressures
with arterial tonometry: clinical revival of the pulse pressure
waveform? Mayo Clinic Proceedings 2010 May 31 (Vol. 85, No.
5, pp. 460–472). Elsevier.
[14] Payne RA, Wilkinson IB, Webb DJ. Arterial stiffness and hyper-
tension emerging concepts. Hypertension 2010; 55: 9–14.
[15] Ring EF, Ammer K. Infrared thermal imaging in medicine. Physi-
ological Measurement 2012; 33: R33.
[16] Schroeder EB, Liao D, Chambless LE, Prineas RJ, Evans GW,
Heiss G. Hypertension, blood pressure, and heart rate vari-
ability the Atherosclerosis Risk In Communities (ARIC) study.
Hypertension 2003; 42: 1106–1111.
[17] Van Bortel LM, Laurent S, Boutouyrie P, et al. Expert consen-
sus document on the measurement of aortic stiffness in daily
practice using carotid-femoral pulse wave velocity. J Hypertens
2012; 30: 445–448.
[18] Wang A, An N, Xia X, Li L, Chen G. A Logistic Regression and
Artificial Neural Network-based Approach for Chronic Disease
Prediction: a Case Study of Hypertension, International Confer-
ence on Internet of Things (iThings 2014), Green Computing
and Communications, 2014.
[19] World Health Organization. Global status report on alcohol and
health. World Health Organization, 2014.
Bereitgestellt von | De Gruyter / TCS
Angemeldet
Heruntergeladen am | 18.07.17 15:56