Rheumatoid Arthritis Detection Using Thermal
Imaging and Fuzzy-C-Means algorithm
Nizami Mohiyuddin
Pradeepkumar Dhage
Krishna K. Warhade
M.E student
Department of Electronics and
Telecommunication
MIT C.O.E. Pune
Assistant Professor
Department of Electronics and
Telecommunication
MIT C.O.E. Pune
Professor
Department of Electronics and
Telecommunication
MIT C.O.E. Pune
nizami.mohiyuddin@gmail.com
pradeep.dhage@mitcoe.edu.in
krishna.warhade@mitcoe.edu.in
ABSTRACT
Rheumatoid arthritis (RA) is a chronic autoimmune disease
which affects the hand joints, wrist, feet, knee, shoulders
and other regions of the body. Even various imaging
modalities like x-rays, CT and MRI are available in
evaluation and diagnosing the disease; those modalities are
expensive and have radiation effects. Thermal imaging
plays a vital role in evaluation and monitoring the
inflammation in rheumatoid arthritis Thermal imaging is a
non invasive method for detecting the pathogenesis of the
disease compared to other diagnostic methods. The
advantage of this imaging technique is that it is a
noninvasive thermographic examination, both from an
operational and health standpoint. The objectives of this
study is to evaluate the rheumatoid arthritis based on heat
distribution index and skin temperature measurements and
to analyze the difference in skin temperature measurement
in various parts of body of RA patients and normal persons.
The algorithm automatically segments the abnormal regions
of the hand especially for arthritis patients using fuzzy-cmeans algorithm and Expectation Maximization (EM
)algorithm.
Infrared imaging is ideally suited to the study of skin
temperature, because the human epidermis has a high
emissivity. This was noted by Hardy, an American
physiologist in 1934 [l]. Sixty years later, with more
sophisticated technology and greater knowledge of
physiology, we still agree with these data. Critics of thermal
imaging point out that the technique only records the skin
temperature. This often stems from a limited and sometimes
outdated understanding of thermal physiology, still taught
from the background of thermocouple recordings in
elaborate laboratory settings. Technical advances in thermal
imaging, particularly since the addition of image processing
techniques have revolutionized the study of skin
temperature. We now know much more about
thermoregulation in man, and the effects of extremes of hot
and cold environment [2]. When inflammation occurs in
deeper tissues and joints, the skin will under the right
conditions show an altered thermal behaviour.
Keywords—Rheumatoid
arthritis,
hand
bone
segmentation, joint margin, automatic joint detection,
thermal imaging, thermography.
I. INTRODUCTION
Rheumatoid arthritis is still a disease of unclear aetiology. It
is a autoimmune disease which causes chronic and
inflammatory disorders and affects the primary joints and it
principally attacks flexible joints. This results in painful
condition which may lead to substantial loss of functioning
and mobility of body. Fig. 1 shows the hand affected with
Rheumatoid Arthritis. Rheumatology embraces a spectrum
of diseases, most of which affect the locomotor system.
Arthritis is a general term which describes articular joint
inflammation. During the inflammatory process the
synovial membrane which supplies lubricant to the joint
becomes thickened and increased blood supply increases
the temperature. In other diseases, such as scleroderma the
circulatory system in the extremities undergoes many
changes, but blood supply is reduced. These and other
rheumatic diseases result in localized changes in
temperature.
Fig. 1 Hand affected with Rheumatoid arthritis
This means that normal control subjects can be used to
establish a healthy baseline, from which patients with
known disease can be compared. This has been achieved in
rheumatology, and good international agreement reached
over the application of thermal imaging. Ultrasound is
dependent on user for imaging, this modality could quantify
changes in synovitis and effusion. For a normal subject,
thermogram shows uniform and symmetrical temperature
variation [3-4]. The assessment of joints with magnetic
resonance imaging (MRI) is too costly and time-consuming
for routine use [5]. Hence Thermal imaging method is
considered as a valuable tool in diagnosing the rheumatoid
arthritis disease. Thermogram depicts a thermal variation in
the skin temperature of various parts of the body But in
case of abnormality condition abnormal areas shows
sudden increase in temperature. The RA affected region
appears as red spot area showing higher temperatures in the
thermogram obtained [6]. To enable earlier diagnosis,
various rheumatological societies have reviewed their
diagnostic criteria and incorporated modern imaging
methods and modalities into their diagnostic algorithms.
II. RELATED WORKS
Many different thermal imaging systems have been tested
over the years since dedicated medical thermographs have
been available Many studies have been performed which
show the anticipated normal pattern of temperature shown
in a thermal image.
Mikhail S. Tarkov et al. have proposed Evaluation of a
Thermogram Heterogeneity Based on the Wavelet Haar
Transform [7]. This method approach is based on a
statistical processing of the thermal image histograms. It is
shown that the histogram transform analysis gives much
new information about change of the human organism state.
At the same time, it is stated that both a sharply
heterogeneous and a sufficiently smooth for visual
perception (diffusive) thermal pictures can give the same
histograms. For this reason, the image heterogeneity degree,
being independent informative characteristic of the thermal
pattern, necessitates a development of special methods for
its quantitative description. The mentioned method devote
efforts to search the quantitative criteria of the image
heterogeneity and adequate algorithms for evaluating the
heterogeneity degree.
Maria del C. Valdes et al. have proposed Multidimensional
filtering approaches for pre-processing thermal images [8].
The method proposed by them effectively corrects some
blurring effects typically found in thermal infrared images.
For the case of a single frame image determines the
direction and width of the blur slope and re-assigns the max
and min values to the correspondent pixels in the gradient
direction. Then, the area is shifted and the same process is
done again, up to cover the full image. Image evaluation
methods demonstrate the accuracy and quality of the results
Christophe L Herry et al. [9] used quantitative assessment
of pain-related thermal dysfunction through clinical digital
thermal imaging. This methods presents methods for
automated computerised evaluation of thermal images of
pain, in order to facilitate the physician'sto make proper
decision. Firstly, the thermal images are pre-processed to
reduce the noise introduced during the initial acquisition
process and to extract the digressive background. Then,
potential regions of interest are obtained using fixed
dermatomal subdivisions of the body, isothermal analysis
and segmentation techniques. Finally, they assess the
degree of asymmetry between contra lateral ROI using
statistical computations.
Mariusz Marzec et al. [10] described automatic method for
detection of characteristic areas in thermal face images.
This paper presents an algorithm for image analysis which
enables localization of characteristic areas of the face in
thermograms. The algorithm is resistant to subjects’
variability and also to changes in the position and
orientation of the head. In addition, an attempt was made to
eliminate the impact of background and interference caused
by hair and hairline. The algorithm automatically adjusts its
operation parameters to suit the prevailing room conditions.
L.A. Bezerra et al. [11] proposed Estimation of breast
tumor thermal properties using infrared images. Firstly is
the development of a standardized protocol for the
acquisition of breast thermal images which includes the
design, construction and installation of mechanical
apparatus. The second part is related to the challenge for the
numerical computation of breast temperature profiles that is
caused by the uncertainty of the real values of the thermo
physical parameters of some tissues. Then, a methodology
for estimating thermal properties based on these infrared
images is presented in the paper.
Carsten Siewert et al. [12] Difference method for analyzing
infrared images in pigs with elevated body temperatures.
The only prerequisite is that there are at least 2 anatomical
regions which can be recognized as reproducible in the IR
image. Noise suppression is guaranteed by averaging the
temperature value within both of these ROI. The
subsequent difference imaging extensively reduces the offset error which varies in every thermal IR-image
The aim of this study was to evaluate and analyse the RA
based on skin temperature differences measurements, and
to automatically segment the abnormal regions of
thermogram using EM algorithm and fuzzy c-means.
II. METHODOLOGY
A. Standardization protocol
Optimal conditions for quantitative thermal imaging have
been published, as consensus reports[13-14]. These
technique which falls within these conditions establishes
the following criteria.
1. Information is supplied to the patient prior to the test, to
avoid major disturbances to the circulation, heart rate or
skin condition. These include smoking, exercise, and
ointments applied to the skin.
2. The patient is briefed, and then rested in a controlled
ambient temperature for a fixed period prior to the test.
Areas to be examined are unclothed, and legs and arms are
stretched out, not crossed during this equilibrium period. A
large chair with arms and a leg rest is ideal for this.
3. The imaging system is calibrated, to an external source if
required, and allowed to run for a period to achieve full
stabilisation. Investigations for inflammatory disorders are
conducted in a 20°C ambient
4. Patients with definite RA (satisfying American
Rheumatism Association criteria) and normal persons
(subjects) were included in this study. The average age of
patient was about 35 years and they were suffering from the
disease for duration of average 6 years. We had taken data
of 4 subjects out of which two are healthy and two are
affected with definite Rheumatoid Arthritis. Consent
statements were signed by each patient.
1
(a) If 𝐼𝑘 =𝜑, then 𝑢𝑖𝑘 𝑏+1 =
B. Thermal imaging process.
∑𝑐𝑗=1(
Imaging was performed at The Centre for Biofield
Sciences, M.I.T College, Kothrud, Pune. Using an infraRent LLC camera (Lakeland, FL 8007099565).The images
were analyzed using MedHot pro IR version 2.0 REV 3
proprietary software. The humidity and air temperature of
the imaging room were maintained stable, with maximum
oscillation in temperature of ±2°C. Thermographic images
ofvarious parts of body were obtained. All thermographic
images are captured at approximately same time of the day
and in the same room.
C. Image processing
maximization algorithm
by
2
𝑑𝑖𝑘 (𝑞−1)
𝑑𝑗𝑘
(3)
)
(b) Else 𝑢𝑖𝑘 (𝑏+1) =0 for all i∉ I and ∑𝑖∉Ik 𝑢𝑖𝑘 𝑏+1 =1; next k.
6. If||𝑢𝑏 − 𝑢𝑏+1 || <∈, stop; otherwise set b=b+1 and go to
step 4.
Thermal image of RA and NRA subjects were captured and
images were converted to HSV and further Fuzzy-c-means
algorithm applied, the segmented image is then
superimposed with original image as shown in Fig 2.
extraction
Thermal image of RA patient
Expectation–Maximization (EM) algorithm is an iterative
method for finding maximum likelihood or maximum a
posteriori (MAP) estimates of parameters in statistical,
where the model depends on unobserved latent variables.
The EM iteration alternates between performing an
expectation (E) step, which creates a function for the
expectation of the log-likelihood evaluated using the
current estimate for the parameters, and a maximization
(M) step, which computes parameters maximizing the
expected log-likelihood found on the E step.
The steps of EM algorithm are as follows.
Conversion of thermal image to HSV
Hue saturation intensity obtained
Fuzzy C means algorithm applied
1. Set K= and initialize θ0 such that 𝐿θk (Y) is finite.
2. Expectation E step: Compute
Q (θ,θ𝑘 ) = 𝐸θk {log 𝑃θk (Z, Y)|Y) = ∫log 𝑃θ (Z,Y)
𝑃θk (Z|Y) dz.
3. Maximization (M) step: Compute
θ k+1 = 𝑎𝑟𝑔θ max Q (θ, θ𝑘 )
Segmented image superimposed with original image
Fig 2. Image segmentation with Fuzzy C Mean
algorithm
4. If not converged update k = k + 1 and return to step 2.
D. Image processing by Fuzzy C Mean Algorithm
The Fuzzy C-Means (FCM) clustering algorithm was first
introduced by Dunn [15] and later was extended by Robert
L [16]. The algorithm is an iterative clustering method that
produces an optimal c partition by minimizing the weighted
within the group sum of squared error objective function
𝐽𝐹𝐶𝑀 [16].
𝐽𝐹𝐶𝑀 =∑𝑛𝑘=1 ∑𝑐𝑖=1((𝑢𝑖𝑘 𝑞 )𝑑2 (𝑥𝑘 , 𝑣𝑖 ))
(1)
A solution of the object function 𝐽𝐹𝐶𝑀 can be obtained via
an iterative process, which is carried out as follows.
1. Set values for c, q and έ.
2. Initialize the fuzzy partition matrix U=[𝑢𝑖𝑘 ]
3. Set the loop counter b = 0.
4. Calculate the c cluster centers 𝑣𝑖 (𝑏) with 𝑢(𝑏)
𝑞
𝑏
∑𝑛
𝑘=1(𝑢𝑖𝑘 ) 𝑥𝑘
𝑏 𝑞
∑𝑛
𝑘=1(𝑢𝑖𝑘 )
𝑣𝑖 (𝑏) =
(2)
5. Calculate the membership 𝑢(𝑏+1) . for k=1 to n, calculate
the following. 𝐼𝑘 = {𝑖|1 ≤ 𝑖 ≤ 𝑐, 𝑑𝑖𝑘 = |(𝑥𝑘 − 𝑣𝑖 )|| = 0} /
𝐼; ; for the 𝑘 𝑡ℎ column of the matrix, compute new
membership values.
III. EXPERIMENTAL RESULTS
A. Skin temperature measurement
It has been observed that the heat distribution in RA subject
is much more than that of the healthy subject. In case of
abnormal conditions the abnormal regions show abrupt
variations in temperature. This variation of temperature is
analyzed for prediction of Rheumatoid Arthritis.
Fig 3 represents thermal image of rheumatoid arthritis
patients showing skin temperature higher in abnormal
regions than the normal regions and also indicates the
region of interest measuring the skin temperature in the
abnormal areas. Fig 4 shows the thermal image of the skin
temperature for a normal patient in an area of interest.
Fig. 3 Increased skin temperature in palm RA patients
Fig. 5(b) Input image of Rheumatoid arthritis subject\
Fig. 4 Skin temperature distribution at palm in normal
individual
After obtaining the HSV image the EM algorithm was
applied
Expectation–Maximization (EM) algorithm is
an iterative
method for
finding maximum
likelihood or maximum a posteriori (MAP) estimates
of parameters in statistical, where the model depends on
unobserved latent variables. The EM iteration alternates
between performing an expectation (E) step, which creates
a
function
for
the
expectation
of
the loglikelihood evaluated using the current estimate for the
parameters, and maximization (M) step, which computes
parameters maximizing the expected log-likelihood found
on the E step. The superimposed image which is obtained
from the final output of non rheumatoid subject is as shown
in the Fig 5(c). The output image of EM algorithm
superimposed after segmentation for rheumatoid subject is
as shown in Fig 5(d). for rheumatoid arthritis subject.
B. Image segmentation results for chest
thermal images
Fig. 5(a) Indicates image of Non rheumatoid arthritis
subject and Fig. 5(b) shows image of rheumatoid arthritis
subject These images were captured by thermal imaging
camera and used for image segmentation process the
images were first converted to HSV image.
Fig. 5(c) Chest NRA image superimposed after
segmentation by EM algorithm
Fig. 5(a) Input Image of Non Rheumatoid subject
Fig. 5 (g) Temperature comparison graph
Fig. 5(d) Chest RA image superimposed after
segmentation by EM algorithm
Fuzzy C-Means algorithm is used for segmentation of the
images which resulted in better output than compared to
EM algorithm. The output images after segmentation are
superimposed by the original input image. The
superimposed image after segmentation by fuzzy-c means
algorithm for non rheumatoid arthritis subject is as shown
in Figure 5(e).
The output after segmentation of the image by FCM
algorithm is superimposed with original image as shown in
the Figure 5(f). for RA subject.
The bar graph in Fig. 5(g) indicates the variation in the
measured skin temperature for rheumatoid arthritis patient
compared to normal participant.
C. Image segmentation results for knee
thermal images
Knee thermal image were analyzed as there is a major joint
knee involved in the leg area which resulted in the rise in
temperature. The thermal images of knee area of
Rheumatoid and Non-Rheumatoid subject were captured.
Rheumatoid Arthritis patient had mentioned pain in this
area so we considered this area in the analysis. The Non
Rheumatoid arthritis subject is as shown in Fig 6(a). The
Rheumatoid subject is as shown in the Fig 6(b).
Fig. 6(a) Knee input image of Non rheumatoid subject
Fig. 5(e) Chest NRA image superimposed after
segmentation by FCM algorithm
.
Fig. 5(f) Chest RA image superimposed after
segmentation by FCM algorithm
Fig. 6(b) Knee input image of rheumatoid arthritis
subject
These images were used as input which had shown
difference in temperature distribution of Non-Rheumatoid
Arthritis subject and Rheumatoid Arthritis patient the knee
area had specially shown the significant increase in
temperature as compared to that of the normal subject. The
EM algorithm was applied to these images for the healthy
subject the EM output image is shown in Fig 6(c). The
image of unhealthy subject after superimposition and
segmentation by EM algorithm is as shown in Figure 6(d).
Fig. 6(f) Knee image of RA subject superimposed after
segmentation by FCM algorithm
The comparison graph shows the difference in temperature
levels of normal and abnormal subject as shown in Figure
6(g). The variation in the measured skin temperature for
Rheumatoid Arthritis patient compared to normal
participant was 1-1.5℃.
Fig. 6(c) Knee NRA image superimposed after
segmentation by EM algorithm
Fig. 6(g) Temperature comparison graph
Fig. 6(d) Knee RA image superimposed after
segmentation by EM algorithm
Fig. 6(e) Knee NRA image superimposed after
segmentation by FCM algorithm
The output of the superimposed image after segmentation
by Fuzzy C-Means algorithm for healthy subject is as
shown in Figure 6(e). The unhealthy subject with knee pain
is as shown in Figure 6(f). The Fuzzy C-Means algorithm
shows good extraction performance.
VI. CONCLUSION
One of the great virtues of this technique is that it is
objective and non-invasive. This means that when the
examination of a patient is difficult, e.g. dealing with young
children or psychosomatic illness, thermal imaging is
particularly useful. In many cases, the technique is not
essential for diagnosis. However, in rheumatology,
monitoring of disease progress is a major concern. In a
disease with no known cure, drug treatment has to be
rigorously assessed. In rheumatic diseases this is not a
simple process. This is borne out by the extensive literature
on the subject and the large number of available tests. No
one single test adequately reflects the complex changes
which occur in the whole patient with an inflammatory
arthritis.. In this paper, we used two segmentation
algorithms like fuzzy-c-means algorithm and EM algorithm
for quantifying and extracting the abnormality of
rheumatoid arthritis patients. The fuzzy clustering
algorithm compares the colors in a relative sense and
groups them in clusters. EM algorithm is an iterative
algorithm of first order so it is slower in convergence. EM
algorithm which is applied for the thermal image processing
of hand region did not provide the accurate and good
results. Rather fuzzy c-means algorithm produced better
results compared to that of EM algorithm.
The need for objective and non-invasive monitoring of
inflammation is therefore ideally met by quantitative
thermal imaging. It is relatively simple and inexpensive,
reproducible under the right conditions, and acceptable to
the patient even when in pain this technology can be used
as a valuable tool for diagnosing the RA patients. Thermal
imaging needs strict protocol, but provides a low cost and
objective tool for non-invasive investigation
VII ACKNOWLEDGMENT
We are thankful to Dr. Aniruddha G. Tembe
Rheumatologist at Aditya Birla Memorial Hospital,
Chinchwad, Pune for his valuable discussion on causes of
RA disease Dr. Aniruddha G. Tembe has provided us the
overview of recent detection techniques used in hospitals
for RA detection and help us in preparation of RA patients
databank. We are also thankful to the Centre for Biofield
Sciences MIT Pune for providing thermal imaging
facilities.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
J.D,Hardy,“The radiation of heat from the human
body”, J Clin Invest, vol. 13, pp. 539-615, 1934.
Y. Houdas, E. F. J. Ring, “Human body
temperature:its
measurement
and
regulation”,
Newyork, Plenum, 1982
N. Selvarasu, “Wavelet based abnormality extraction
and quantification algorithm for thermographs
depicting diseases in human”, International
Conference on Fiber Optics and Photonics. December
13-17, 2009, IIT Delhi, India.
Yinghe Huo, Koen L. Vincken, Max A. Viergever,
Floris P. Lafeber “Automatic joint detection in
rheumatoid arthritis hand radiograph”, In IEEE 10th
International Symposium on Biomedical Imaging:
From Nano to Macro San Francisco, CA, USA, April
7-11, 2013.
P. T. Kuruganti and H. Qi. “Asymmetry analysis in
breast cancer detection using thermal infrared images”.
In Proc. Of the SPIE, vol. 5959, pp. 147-157, 2005.
Syaiful Anam, Eiji Uchino, Hideaki Misawa, and
Noriaki Suetake. “Automatic bone boundary detection
in hand radiographs by using modified level set
method and diffusion filter”, In IEEE 6th International
Workshop on Computational Intelligence and
Applications, Hiroshima, Japan, July 13, 2013.
J. Mikhail S. Tarkov, Boris G. Vainer Horvath,
“Evaluation of thermogram heterogeneity based on the
wavelet haar transform”, In Siberian Conference on
Control and Communications SIBCON-2007.
Maria del C. Valdes, Minoru Inamura, J. D. R. Valer,
Yao Lu,”Multidimensional filtering approaches for
pre-processing thermal images”, Multidim Syst Sign
Process, vol. 17, pp. 299-325, 2006.
Christophe L. Herry, Monique Frize, “Quantitative
assessment of pain-related thermal dysfunction
through clinical digital thermal imaging”, In
BioMedical Engineering Online, pp. 3-19, 28th June
2004.
[10] Mariusz Marzec, Robert Koprowski, Zygmunt Wrobel,
Agnieszka Kleszcz, Sławomir Wilczynski, ”Automatic
method for detection of characteristic areas in thermal
face images”, In multiple tools appl, vol. 13, pp. 145149, 2013.
[11] L.A. Bezerra, M. M. Oliveira, T. L. Rolim, A. Conci,
F. G. S. Santos, P. R. M. Lyra, R. C. F. Lima.
“Estimation of breast tumor thermal properties using
infrared images”, In signal processing journal,
springer, vol. 93, pp. 2851-2863, 2013.
[12] Carsten Siewert, Sven Danicke, Susanne Kersten,
Bianca Brosig, Dirk Rohweder, Martin Beyerbach,
Hermann Seifert, “Difference method for analysing
infrared images in pigs with elevated body
temperatures”, In Z. Med. Phys, vol. 24, pp. 6-15,
2004.
[13] J. M. Engel, J. A. Cosh, E. F. J. Ring, "Thermography
in locomotor diseases recommended procedure", Eur.
J. Rheum. Inflamm. vol. 2, pp. 299-306, 1979.
[14] E. F. J. Ring, J. M. Engel, D. P. Page Thomas,
“Thermologic methods in clinical pharmacology", Int.
J. Clinical Pharmacology, Therapy and Toxicology,
vol. 22 no. 1, pp. 20-24, 1984.
[15] Dunn J. C. “A fuzzy relative of the ISODATA process
and its use in detecting compact well separated
clusters”. Journal of Cybernetics, vol. 3, 1974, pp. 32–
57.
[16] Robert L. Cannon, Jitendra V, “Efficient
implementation of the fuzzy c-means clustering
algorithms”, IEEE transactions on pattern analysis and
machine intelligence. vol. PAMI-8, no. 2, March 1986.