Journal of Medical and Biological Engineering, 22(2): 75-82 75

The Application of Thermal Image Analysis to Diabetic

Foot Diagnosis
Kuo-Sheng Cheng* Jin-Shen Yang Ming-Shi Wang1 Shin-Chen Pan2
Institute of Biomedical Engineering, National Cheng Kung University, Tainan, Taiwan, 701, ROC
1
Department of Engineering Science, National Cheng Kung University, Tainan, Taiwan, 701, ROC
2
Department of Plastic Surgery, National Cheng Kung University Hospital, Tainan, Taiwan, 701, ROC

Received 3 May 2002; Accepted 12 June 2002

Abstract
Due to the deficiency of peripheral sensory and autonomic nerves and the complication of peripheral vascular
disease, the patients with diabetes are prone to induce the ulcers on their limbs. If the ulcer formation or poor
circulation area can be detected as early as possible, the occurrence of amputation morbidity may be reduced. In this
study, the thermal image analysis is applied to characterize the superficial circulation of foot. Nine subjects with or
without diabetes needing skin graft are recruited and divided into control group and experimental group. Both static
and dynamic thermal images are acquired for analysis. The temperature difference percentage, Karhunen-Loéve
transform, and thermal recovery tendency analysis are employed for thermal image analysis. From the results, it is
shown that the area of poor circulation around the wound in the patients with diabetes may be significantly revealed
using Karhunen-Loéve transform and thermal recovery tendency analysis. The thermal recovery tendency analysis
needs shorter computation time and is a good parameter to early detect the ulcer formation.

Keywords: Thermal image, Diabetic foot, Skin graft

Introduction
image [5]. For patients with diabetes mellitus, the frequently
Human has a good homeostatic control system for body occurred complication is lower extremity amputation. It is
conditions, including temperature, pH, etc. The temperature due to foot ulceration that results in osteomyelitis, sepsis and
is a useful parameter for revealing the body healthy peripheral tissue necrosis. The ulceration almost begins with
condition. Normally, the core temperature in our body is poor circulation area. Therefor, foot screening regularly is
maintained at the 36.8 ± 0.6 degree of centigrade. In the very important for patients with diabetes. If the poor
extremity of body, temperature is especially a good indicator circulation area may be detected as early as possible, the
to reflect its condition. If the temperature within some area morbidity of lower extremity amputation may then be
becomes higher or lower than other areas unnaturally, then it reduced [6-9].
will be considered to have some problem such as infection, In 1993, Chen developed the relationship between blood
necrosis, etc. Generally, the core body temperature can be flow and surface temperature [10]. Based on one-dimension
easily measured by using mercury thermometer or electronic simplification of the heat transfer from arteriole to dermis,
thermometry. But in the extremity, the temperature dermis to epidermis, and epidermis to the surroundings, a
distribution is widely varied and hardly measured using linear relationship between blood flow and surface
traditional thermometry. temperature for the skin may be obtained and demonstrated
Infrared thermal image has been applied in many fields, with a physical model. Later, Wu [11] applied the thermal
such as military, industry, and medical field. Recently, the imaging technique to screen the small vascular disease. From
infrared thermal imaging technique is reapplied to breast the fractal dimension of static thermal images, it is possible to
cancer detection, peripheral arterial disease diagnosis, differentiate the patients with diabetes from normal patients.
osteoarthritis diagnosis, and pain evaluation [1-4]. Infrared In 1998, Mabuchi et al. proposed that thermal distribution of
thermal imaging was employed for detection of chronic the human body based on the dermatome distribution should
neuro-muscular injury that may not be seen in CT or MRI be symmetry [12]. For the joint arthritis patient, it is a good
parameter for diagnosis. However, it is not suitable for the
* Corresponding author: Kuo-Sheng Cheng patients with diabetes, because all the vessels in body may be
Tel: +886-6-2757575 ext. 63424; Fax: +886-6-2343270
affected and hardly be differentiated between normal side and
E-mail: kscheng@mail.ncku.edu.tw
76 J. Med. Biol. Eng., Vol. 22. No. 2 2002

affected side. In 1987, Unser et al. analyzed the dynamic area of the hypothalamus serves as the primary overall
sequences of thermal images for detecting breast cancer using integrator of the reflexes, but other parts of brain also serve
Karhunen-Loève transform [13]. Due to angiogenesis, the as some control over specific components of the reflexes.
temperature in cancerous area usually is higher than normal Output from the hypothalamus and the other brain areas to
one. The tumor area showed less temperature changes than the effectors are through two pathways: (1) Sympathetic
the nearby area. nerves to the sweat glands, skin arterioles; and (2) motor
In summary, the purposes of this study are in two folds. neurons to the skeletal muscles.
First, it is to develop the thermal image analysis method for Changes in skin blood flow alone can regulate body
detecting and locating the poor circulation area. Second, it is temperature by altering heat loss over a range of
to develop the friendly user interface of the thermal image environmental temperatures (approximately from 25 to 30
analysis system for clinical application. degrees in centigrade). It is known as the thermoneutral zone.
When the enviromental temperature above the thermoneutral
Materials and Methods zone, the sympathetic nerves induce the arterioles to
vasodilate to regulate the excess amount of heat. As the
Background
maximal vasodilation can’t eliminate heat as fast as it is
Infrared thermal image is an image that captured by a
produced, and another heat loss mechanism, sweating, is
thermal camera to display the infrared radiation. All the
therefore brought strongly into play. For the motor nerve, it
objects with temperature above absolute zero will emit
induces the skeletal muscles to relax to avoid too much heat
electromagnetic radiation spontaneously. By definition, all
production.
the incident radiations will be absorbed and then radiate by a
Basically, these processes will be seen in the person
black body in a continuous spectrum accounting to Planck’s
whose sensory organ (peripheral thermoreceptors), the nerve
law. Integrating Planck function over all frequencies leads to
function and arterioles are intact. For the patients with this
the Stefan Boltzmann’s law for the total emissive power from
kind of disorder, some of the control process may be blocked.
a black body as E b = σT 4 . σ is the Stefan-Boltzmann’s
Thus, when the heat is applied to skin of subject for 3 min,
constant, and T is absolute temperature. Furthermore, from
and then removed. The tendency of temperature changes for
the Wien’s law, the emissive wavelength of maximum power
the healthy subjects and the diabetic patients will be very
is dependent on the temperature in dimensionless
different.
representation as
Image Analysis
2898
λ max = The thermal image of foot sole is acquired for diagnosis.
T (1) In this study, three methods are employed to analyze the
where λ max is the wavelength (in µm) of the maximum thermal images. Firstly, the ROI is selected manually on
emissive power at temperature (in K). For example, radiation monitor by defining four feature points with mouse. These
from the human body at 310 K is maximal at a wavelength of points are the position besides first and fifth metatarsal head,
9.3 µm, and it is in the infrared region. Therefore infrared and both sides of the calcareous. The major problem for the
thermal image is a suitable tool to measure the human skin image sequence taken along the time is that part of ROI may
temperature [14], and indirectly related to superficial blood be subject to small displacements from one time frame to
flow. another due to accidental movement of the patient. Thus, the
Physiology of Thermal Regulation images need to be corrected using a geometric transformation
before subtraction. A geometric transformation is a transform
Basically, the blood vessels in the skin are under the
that maps the pixel ( x, y ) in one image to a new position
control of nervous system. The sensory, motor, and
sympathetic nerve innervates human peripheral nerves in
( x' , y ' ) in another image. From the defined feature points, it
may be easily obtained using affine transform.
limbs. When the ambient temperature changes, the body
temperature may be regulated according to physiological Spatial-Temporal representation is used to represent the
mechanism [15]. thermal image variation with time [16, 17]. Consider a time
series of 600 thermal image (about 8000 pixels) taken for the
In human skin, several kinds of sensory receptors
foot’s response to the thermal stress and let Pi be the
present, and thermal receptor is one of them. There are two
temperature vector of corresponding pixel i in the 600
categories of thermoreceptors, one in the skin (peripheral
thermal images, for l = 1,...,600 and j = 1,...,8000 , we have
thermoreceptors), and the other (central thermoreceptors) in T

deep body structures, including the hypothalamus, spinal cord,  

P1 =  t 11 , t 21 ,..., t
600 1 
and abdominal organs. When the environmental temperature  
changes, the thermal receptor will be activated and induce a  
T

P2 =  t 12 , t 22 ,..., t
series of reactions. The central chemoreceptors provide the 600 2 
 
essential negative-feedback component of the reflexes to ·
maintain the body temperature. The peripheral chemore-
·
ceptors provide feedforward information, and also account T
  (2)
for one’s ability to identify a hot or cold area of the skin. An P8000 = t ,t ,..., t 
 1 8000 2 8000 600 8000

 Thermal Image Analysis 77

the images of spatial-temporal representation form are fitted

using Eqs. (3) and (4).

y = ln( x / τ + a) + a0 (3)
Temperature Changes (℃)

y = a (1 − exp − x / τ ) + a0 (4)
The time constant τ of each pixel of images may be displayed
for comparison. An example of fitting curves is shown Figure
1.
Graphic User Interface Design
Through the GPIB interface card, we can control the
thermal-imaging device by personal computer. It will be
Times (sec) convenient for clinical use. The configuration of user
interface is shown in Figure 2.
Figure 1. A point temperature changes with time (－) with nature log
The thermographic analysis software and user interface
fitting function(－) and exponential fitting function (－).
is developed with Microsoft Visual C++ 6.0, and executed
under Microsoft Windows 98.

Experimental Design
Nine patients with or without diabetes are recruited for
this study. The age ranges from 18 to 81. They are divided
into experimental group and control group. In experimental
group, the subjects are the inpatients with the diagnosis of
diabetes mellitus for 10 to 20 years excluding the
hypertension or cardiovascular disease and have ulcers on
their feet needed surgery to debride and skin graft. In the
control group, they are the patients excluding hypertension or
cardiovascular disease whose feet suffered from trauma
needed skin graft.
The infrared imaging system used in this study is San-ei
Thermo Tracer, Th2100 (NEC Co.). The wavelength for input
measurement is 8 to 13 µm. The temperature resolution is
Figure 2. GUI of thermal image acquisition system 0.02 oC, the image size is 256 by 191 pixels, and the frame
rate is 20 Hz. The personal computer (Pentium III) with the
GBIP is used as control unit for acquiring, processing, and
Temperature Difference Percentage
storing the thermal images.
For static thermal image analysis, 20 images before and
Image Acquisition
after surgery without the application of thermal stress are
All the experimental procedures can be divided into
acquired, respectively. 20 images are averaged to obtain the
static and dynamic image acquisition. In the static acquisition
mean temperature. Then the difference between the mean
procedure, the subjects are asked to supine on the bed. The
temperatures before and after surgery are calculated and
thermal camera focuses on the subject’s feet and captures the
scaled for gray level display.
thermal images continuously for 60 seconds. The image
Karhunen-Loève Transform
sampling rate is 1 Hz. In the dynamic acquisition procedure,
Karhunen-Loève transform (KLT) is used for extracting a light as the heat source is applied to increase the skin
the principal temperature component for analysis [18]. The temperature up to 2 centigrade degrees. This procedure needs
“eigen-images” before and after surgery corresponding to the about 3 to 5 minutes. Then, the thermal camera focuses the
largest eigen value are then displayed for comparison. The subject’s feet and captures the thermal image continuously
resultant eigen-images are used in the next stage of diagnostic for 600 seconds. The image sampling rate is the same as
analysis. static acquisition procedure. Those image data are acquired
Thermal Recovery Tendency from the infrared imaging device by GPIB interface and
As mentioned above, the tendency of thermal regulation saved in hard disk. The whole system configuration of the
may reflect the vascular state for diagnosis. From the experiment is depicted in Figure 3.
dynamic thermal images, the tendency of temperature Environment Control
changes is used a new parameter for characterizing the blood The experimental condition control is very important in
perfusion of skin. The temperature changes of some point in this study. There are some points that may disturb the thermal
78 J. Med. Biol. Eng., Vol. 22. No. 2 2002

Figure 3. System configuration for experiment

>5 %

good

No change fair

poor
< -5 %

(a) (b)
Figure 4. (a) The gray scale representation for temperature difference percentage, and (b) The gray scale representation for thermal recovery
tendency analysis and Karhunen-Loeve transform analysis.

image data during acquisition. Standardized ambient dynamic thermal images, 10 gray levels are used to present
conditions are essential for thermal imaging. Skin surface the temperature difference percentage as displayed in Figure
temperature is determined by the rate of heat exchange 4. The more the temperature drops, the darker the gray level
between the surrounds and the body [9]. It is important to will be.
keep the ambient temperature at a constant level, and the air Temperature Difference Percentage
flow below 0.2 m/sec [19]. Typically, the subject is asked to The images of temperature difference percentage for
relax and lie on bed for about 20 min in the examination control group and experimental group are shown in Figures 5
room and the infected part is exposed to the ambient so that and 6, respectively.
thermal equilibration takes place. The room temperature is set Thermal Recovery Tendency
at within 22~23℃ (22.7 ± 1.6℃), and the humidity is keep in
The images representing the time constant of thermal
the range of 52~68% (57.9 ± 4.8%). For reducing the effect
recovery tendency for control group of subject 1 and
of air circulation, the subject’s lower extremity and infrared
experimental group are shown in Figures 7 and 8,
camera are both covered in a plastic box.
respectively.
Results Karhunen-Loève Transform
Figure. 9 shows the first three eigen images of subject 1
In the resulting images shown in this section, a square
before and after surgery, respectively. In this study, only first
mark (□) is used to point out the observation window for the
eigen-image (the principal component) with the largest eigen
region around wound. The arrowhead ( ) points an area
value containing the most feature is used for comparison.
with revealed pathological blood circulation. In the static and
 Thermal Image Analysis 79

(a) (b) (c)

Figure 5. The images of temperature difference percentage for control group of (a) Subject 1, (b) Subject 2, and (c) Subject 3

(a) (b) (c)

(d) (e) (f)

Figure 6. The images of temperature difference percentage for experimental group of (a) Subject 4, (b) Subject 5, (c) Subject 6, (d) Subject 7, (e)
Subject 8,and (f) Subject 9

(a) (b)
Figure 7. First eigen-image for Subject 1 of control group (a) before surgery and (b) after surgery
80 J. Med. Biol. Eng., Vol. 22. No. 2 2002

(a) (b)
Fig. 8 First eigen-image for Subject 4 of experimental group (a) before surgery (b) after surgery

(a) (b) (c)

(d) (e) (f)

Figure 9. First three eigen-image of Karhunen-Loève transform for subject 1 of control group. (a) and (b) are the first eigen-image before surgery
and after surgery, (c) and (d) the second eigen-image before surgery and after surgery, and (e) and (f) the third eigen-image before
surgery and after surgery

Table 1. The information for experimental subjects the area around the wound or observation window will reveal
poorer circulation after skin graft surgery. From the resulting
Experimental group Control group
images, it is demonstrated in the experimental group but not
6 (Subject 4 – 9) 3 (Subject 1 – 3)
Number in the control group.
49~81 (63.5±26.82) 18~77 (45.33±42.06) From the resulting images of temperature difference
Age
F: 2 M: 4 F: 2 M:1 percentage, it is seen that the temperature drops in Subject 4,
Gender
7, 8, and 9, not obviously. The temperature rises or retains in
Diabetic mellitus for Patients without diabetic
History subject 2 and 3. This method is not very effective in diagnosis
more than10 years and mellitus and with foot
of poor circulation area. The point chosen for temperature
with foot ulceration trauma
adjustment in computation of temperature difference is
Discussion important in this method. It needs further study.
From the resulting images obtained by thermal recovery
In this study, all patients received skin graft surgery. The
tendency analysis, it is seen that the temperature drops
skin graft area has smaller arteriole resistance. Due to
significantly in Subject 4, 6, 7, 8, and 9. The temperature
microvascular stealing phenomenon and nerves and vessels
rises or retains in Subject 1, 2, and 3. This method is feasible
disorder, the areas around wound will not have temperature
to screen the poor circulation area.
changes during the cooling process. Thus, it is expected that
 Thermal Image Analysis 81

Based on KL transform, The resulting images of first [2] E. F. J. Ring, “Progress in the measurement of human body
eigen images obtained before and after surgery are temperature”, IEEE Eng. Med. Biol., 17(4):19-25, 1998.
[3] B. F. Jones, “A reappraisal of the use of infrared thermal image
demonstrated to be very effective in diagnosis of circulation analysis in medicine”, IEEE Trans. Med. Image, 17(6):1019-
problem in thermography. It is seen that the temperature 1027, 1998.
drops are very significantly in Subject 4, 6, 7, 8, and 9. The [4] J. R. Harding, D.F. Wertheim, and R. J. Williams, “Infrared
temperature rises or retains in Subject 1, 2, and 3. However, it imaging in diabetic foot ulceration”, Proc. 20th conf. IEEE
is very intensive in computation. Eng. Med. Bio. Soc., 20(2):916-918, 1998.
[5] B. Harrison and K. Mabuchi, “Biomedical infrared imaging in
In the experimental procedure, besides environment Japan”, IEEE Eng. Med. Biol., 17(4):66-70, 1998.
control, and subject’s limbs fixation are important. It will [6] D. G. Armstrong and L. A. Lavery, “Diabetic foot ulcers:
arise some difficulty in the image registration for temperature prevention, diagnosis and classification”, Amer. Famli. Phys.,
difference computation. 57(6):1325-1332, 1998.
In this study, only nine subjects are analyzed with part in [7] M. E. Cooper, R. E. Gilbert, and G. Jerums, “Diabetic vascular
complications,” Clin. Exp. Pharma. Phys., vol. 24, pp. 770-775,
quantitative. It needs to collect more data for improving the 1997.
method in the future. [8] S. P. O’Beien, M. Schwedler, and M. D. Kerstein, “Peripheral
Although Karhunen-Loève transform is a better way to neuropathies in diabetes”, Surg. Clin. North Amer., 78(3):393-
extract the area of poor circulation and early ulceration, but it 408, 1998.
will spend a lot of time to compute the eigen-image. The [9] C. J. McCabe, R. C. Stevenson, and A. M. Dolan, ”Evaluation
of a diabetic foot screening and protection program”, Diabetic
thermal recovery tendency is also a good method to observe Med. 15:80-84, 1998.
the abnormal area, and for each pixel, it only needs to use the [10] J. Y. Wu, Thermogram Analysis and Its Clinical Application,
curve fitting to find the thermal recovery tendency. For Master Thesis, National Cheng Kung University, Tainan,
example, the KLT will take about half an hour to process 1 Taiwan, ROC., June 1995.
subject’s data, but the thermal recovery tendency method will [11] H. F. Chen, The Application of Fractal Dimension to Analyze
Cutaneous Blood Perfusion, Master Thesis, National Cheng
only spend 2 min to compute 1 subject’s data. Kung University, Tainan, Taiwan, ROC., June 1993.
[12] K. Mabuchi, T. Chinzei, and I. Fujimasa, “Evaluating
Conclusion asymmetrical thermal distributions through image processing”,
IEEE Eng. Med. Biol., 17(4):47-55, 1998.
In this study, few points may be summarized as follows. [13] M. Unser, H. V. Hamme, and P. D. Muynck, “Karhunen-Loève
First, the cooling process is a feasible method for the analysis of dynamic sequences of thermographic images for
diagnosis of the pathological blood circulation area or early early breast cancer detection”, Proc. Conf. CVPR, 592-596,
1998.
ulceration. Second, the Karhunen-Loève transform is a good [14] K. S. Ghosh and P. E Galeski, “Criteria for selection of
way to reveal the pathological area. However, it is very infrared camera system”, Proc. Conf. Industry Appli.,3:1893-
intensive in computation. Third, thermal recovery tendency 1897, 1994.
method is proposed to find the tendency in the cooling [15] A. J. Vander, J. H. Sherman, and D. S. Luciano, Human
process using regression technique. It is very fast in Physiology, 6th edition, McGraw-Hill, Inc., 1994.
[16] M. J. Varga and P. D. Muynck, “Thermal analysis of infra-red
characterize the dynamic of temperature changes of skin, and mammography”, Proc. Conf. IAPR , 360-364, 1992.
demonstrated to be a novel method for detecting the [17] M. J. Varga and R. Hanka, “Automatic thermal image analysis
pathological blood circulation area. for medical diagnosis”, Proc. Conf. Image Processing and Its
Applications, 526-529, 1992.
References [18] R. C. Conzalez and R. E. Woods, Digital Image Processing,
Cambridge, MA: Addison-Wesley, Ch. 31:48-156, 1992.
[1] I. Fujimasa, T. Chinzei, and I. Saito, “Converting far infrared [19] I. D. Swain and L. J. Grant, “Methods of measuring skin blood
image information to other physiological data”, IEEE Eng. flow”, Phys. Med. Biol., 34(2):151-175, 1989.
Med. Biol., 19(3):71-76, 2000.