ISSN XXXX XXXX © 2016 IJESC
Research Article
Volume 6 Issue No. 8
Surface Temperature Distribution in Popliteal Region for Early
Detection of Osteoarthritis
Aishwarya Jairam1 , Joshi Manisha S2 , Dr. V Umadevi3
Depart ment of Medical Electronics 1, 2 , Department of Co mputer Science 3
BMS Co llege of Engineering, Bangalore, India
Abstract:
Infrared thermography (IRT) is used to measure and analyze physiological functions and pathology which can be related to the
body’s thermal ho meostasis and temperature. Our goal is the early detection of Osteoarthritis (OA), with a narrow focus on skin
surface temperature distribution around Popliteal (Posterior knee) region. The the temperature distribution along the popliteal
region was considered whereas previous works focused on the temperature elevation in the patellar reg ion affected with OA [13 15]. The average temperature was found to be variable for different person. So it is difficu lt to detect early OA knee considering
temperature elevation instead here temperature distribution in popliteal region is considered. Infrared thermography and
radiography were done on 27 normal knees (Kellgren-Lawrence grade 0) and 34 early Osteoarthritis knee (Kellgren-Lawrence
grade 1). When a vertical profile plot of popliteal region was considered surface skin temperature was found to be more
distributed in entire region for early OA knee and peaked near knee jo int for normal knee. Statistical measures that predict the
temperature distribution in popliteal region were fed to a support vector machine (SVM) classifier to perform automated
recognition of early OA knee. Experimental results indicate that the SVM classifier has an average accuracy rate of 86.65%, a
sensitivity of 84.42%, and a specificity of 88.89% in detecting normal and early osteoarthritis cases. The proposed method for
knee thermal screening can thus provide quantitative reference in formation in assisting clinical diagnosis.
Keywords: Popliteal region, Infrared Thermography, Rad iography, Surface temperature distribution
I.
INTRODUCTION
Osteoarthritis is a progressive form of arthrit is which is
characterized by breakdown of the cartilage in joints [1].
Osteoarthritis is abbreviated as OA and is also known as
degenerative joint disease. OA will cause pain in the joints
with activity. The knees and the hips are common jo int
locations affected with OA and are common in people over 60
years of age, but it can affect younger people when they have
had joint inju ry or surgery [2].
Our goals were to standardize a method of knee infrared
imaging to explore which distinct knee region provides the
highest correlation of local temperature distribution with early
radiographic findings of knee OA and to characterize
osteoarthritis disease at early stage using Infrared
Thermography which will allow a paradig m shift fro m
palliation of late disease towards prevention. The early
diagnosis and early treatment of joint in jury and degeneration
reduces osteoarthritis risk.
II.
OA develops over years, offering a long window of time to
alter its course potentially. The etiology of OA shows strong
associations with highly modifiab le risk factors such as
mechanical overload, obesity and joint injury. Osteoarthritis in
its late-stage, modifying disease opportunities is limited.
Radiographic joint-space narrowing is highly insensitive
measure used for centuries to diagnose OA. Advances in
optical imaging and magnetic resonance imag ing (MRI) have
permitted direct imaging of joint tissues that helps in
quantitative assessment of joint tissue structure for identifying
early OA [3]. But MRI require patient to stay still for a long
time inside a machine. This can be uncomfortable. X- Ray and
CT involve exposure to a small amount of radiation. IRT
belongs to the functional imaging category; it provides a noninvasive and dynamic measure of the heat radiated fro m the 1–
2 mm belo w the skin surface [4].
A. Subjects
Infrared thermal imaging has been used in a range of clinical
applications in rheu matology including: the monitoring of skin
temperature elevations for pain assessment [5-7], evaluation of
rheumatoid arthrit is (RA) [8-11] hand OA [11,12] and knee
OA[13-15], animal clinical trials[16].
B. Knee Radiography
International Journal of Engineering Science and Computing, August 2016
METHODS
54 subjects participated in this study. An Arthritis Camp was
organised on 30th Jan 2016 in association with Medical
Electronics Depart ment of BMS Co llege of Eng ineering and
Sagar Clin ic (Banashankari, Bengaluru, India ) and funded by
Teqip Phase II. The entry criterion for OA participants chosen
included definite x-ray evidence of knee OA of KellgrenLawrence (KL) [17] grade 1,2 3 or 4 chosen on the basis of xray. Controls were required to have normal (KL0) . As this
study is about early diagnosis of OA, only knee with KL grade
0 and 1 was considered
Participants were informed in detail of the study protocol,
signed a consent statement and could at any time if they
wanted to abort the protocol.
Posteroanterior digital x-rays were obtained at Sagar Clinic
(Banashankari, Bengaluru, India). X-rays were read by a
single radiolog ist (Dr. Srinivasmurthy) for ev idence of OA
using the Kellgren Lawrence (KL) grad ing system with a
standard paper [17]: grade 0= No evidence of OA(normal);
2216
http://ijesc.org/
grade 1=early OA changes noted(doubtful joint space
narrowing); grade 2=OA changes noted(definite jo int space
narrowing); grade3 = OA changes with narrowing of intra
condylar joint(moderate OA); grade 4= advanced OA changes
with narrowing of intra condylar jo int(severe OA)
C. IRT Procedure
The thermal image was taken using KT 640 fro m Sonel Test
and Measurement Instruments. It is fully radio metric camera
which records temperature at each point of the image -high
resolution 640 x 480. Record ing is in the extended jpg format.
Infra Fusion technology of the camera used to view co mbined
real and infrared image for effect ive locating of the
measurement place.
The IR camera captures the emitted radiat ion. Therefo re the
consideration of the environmental condition is very crit ical if
not the temperature measurement of the target object may be
corrupted from other infrared sources. So the environment
where the IRT was done had limited source that emits IR
radiation and the room temperature was maintained at 24°C.
All subjects underwent IRT in same roo m. Part icipants got an
hour to rest in a comfortable position after X-Ray before IRT
was conducted to ensure that the participants did not have any
stressful physical activity that may increase their body
temperature. During the camp no caffeine (tea or coffee) was
served that may add to the body temperature instead non
caffeine drinks were served. Infrared thermal images of the
knees were acquired in four configurations (anteroposterior,
posteroanterior, latero medial, and med iolateral). The
posteroanterior image was used for the analysis.
D. Image Preprocessing
The real image and thermal images obtained from the Sonel
IR ca mera were not correlating with each other. For this
purpose a Sub-Image matching algorith m was developed to
match the real image with the thermal image.
1.
2.
3.
4.
Both IR and Visual image are converted to grey
scale(Figure 1)
Edge detection of both IR and Visual image using Can ny
edge detection. Visual Images edge detection was done
with higher sig ma value to get a better correlation
between two edge detected image (Figure 2)
All possible sub-image of Visual edge detected is
correlated with the edge detected IR image using Pearson
Correlation.
The Coordinates of sub-image with highest correlation
value is used to crop the Visual image and resize to 640 x
480(Figure 3)
Figure 1: Grey scale image of IR Image (Left) and Visual
Image (Right)
International Journal of Engineering Science and Computing, August 2016
Figure 2: Canny Edge detection of IR i mage (Left, No
smoothing is done i.e. Sigma=0) and Visual Image (Right,
Smoothing is done i.e. Sigma=5)
Figure 3: Image Overlap by al pha blending techni que.
Before appl ying Sub-Image matching algori thm (Left).
After appl ying Sub-Image matching algorithm Right
E. Automatic selection of ROI
For automatic selection of ROI the centre of popliteal region
in real image was manually marked as s een in top left corner
of Figure 5. Following are steps involved in the ROI
extraction:
1.
2.
3.
Separation of left and right posterior knee
Identify the Popliteal centre
Extract 30X90 rectangular region around the popliteal
centre identified in previous step
Steps for separating left and right posterior knee:
1.
2.
3.
4.
5.
IR image is converted to grey scale and segmented using
otsu method
The profile change along the x axis at 60% (288th pixel) of
Y axis is dig italised as shown in Figure 4
All the pixels at which the profile change is detected are
saved as a list.
The middle point of 2nd and 3rd pixel stored in previous
step is used as the crop axis for separating the two knees .
The real and IR image are cropped at crop axis estimated
in previous step. The final result is shown in Figure 5
Figure 4: Profile Change between two legs along x axis on
60% (288 th pixel) of Y axis
2217
http://ijesc.org/
elevation instead here temperature distribution in popliteal
region is considered. The Figure 7 and Figure 8 shows the IR
image and vertical profile plot along the popliteal region
selected for both normal and early OA subject respectively.
For normal knee the vertical profile plot is peaked near knee
joint while in the plot of the early OA knee temperature is
more distributed in the entire popliteal reg ion. Therefore
statistical measure that relates to the temperature distribution
in the entire popliteal reg ion is extracted as attributes to the
classifier model in next section.
Figure 5: IR Image and Visual Image of the separated
knee
After left and right knee separation, the manually marked
popliteal centre is to be identified.
Steps to identify posterior knee (Popliteal) region:
1.
2.
3.
4.
The cropped visual image of right knee/left knee is
converted to grey image and segmented using otsu
method
The profile change along the x axis for every pixel in y
axis is digitalised. (Figure 6)
When 3 or more p rofile change is detected, all the pixels
at which the profile change is detected are saved as a list
Pixel between 2nd and 3rd profile change is stored as
posterior knee centre
Figure 7: Thermal Image of Normal Subject wi th left
popliteal region selected (Left). Vertical Profile Plot of the
popliteal region selected (Right)
Figure 8: Thermal Image of early OA subject wi th left
popliteal region selected (Left). Vertical Profile Plot of the
popliteal region selected (Right).
G. Feature Extraction and Classification
Attributed considered for classificat ion are the s tatistical
features extracted fro m the selected ROI i.e. the popliteal
region. Statistical measures considered are:
1.
2.
3.
4.
5.
6.
7.
8.
Figure 6: Profile change when knee marker is not detected
(Top). Profile change when knee marker is detected
(Bottom).
After the identificat ion of the posterior knee centre a
rectangular region (30X90) around the popliteal centre is
selected and saved fro m wh ich features are ext racted.
F. Image Analysis
Here the focus was on the temperature distribution along the
popliteal reg ion which is the region of interest (ROI) whereas
previous works focused on the temperature elevation in the
patellar region affected with OA [13-15]. The average
temperature was found to be variable for d ifferent person. So
it is difficult to detect early OA knee considering temperature
International Journal of Engineering Science and Computing, August 2016
Variance
Standard deviation
Average absolute deviation
Skew (3rd order mo ment about the mean)
Kurtosis(4th order mo ment about the mean)
Difference between Minimu m and Maximu m p ixel value
Difference between Maximu m and Median pixel value
Difference between Median and Minimu m pixel value
The work here considers the grey values in feature extraction.
It was found that the grey pixel value is directly proportional
to temperature pixel value
Among the classification methods available for solving pattern
recognition problems, the SVM classifier has been proven to
be an efficient way of obtaining good classification
performance [18]. Therefore, the features extracted are fed to
the SVM classifier to develop an automated early diagnostic
method for knee OA.
The process of SVM classificat ion involves two primary
procedures, namely those of training and testing. Based on the
training dataset (feature vector and class label contained in
each dataset), the goal of the SVM classifier is to produce a
decision model that can predict the class label of the test data
2218
http://ijesc.org/
given only the feature vector. The training data and test data
was split into 75% and 25 % ratio respectively. The Input
details of SVM classification is given in Table 1.
Table 1: Input for SVM Classification
Number
of
attri butes
8
III.
Number
instances
61
of
Fro m the results we can conclude that the temperature
distribution in popliteal region significantly helps in diagnosis
of knee OA at its early stage (i.e. radiographic findings with
KL grade 1). .
COMPETING INTERESTS
Associated Tasks
All authors declare that there is no conflict of interest.
Classification
Class 0: Normal knee
Class 1: Early OA knee
ACKNOW LEDGEM ENTS
RESULTS AND DISCUSSION
A total of 27 normal knees (KL 0) and 34 early OA knees (KL
1) were selected on the basis of Radio graphical findings. The
majority of subjects were wo men (55%). Mean age was
48.675 year (range 19–68 year). The each knee was evaluated
thermally and radiographically. Kurtosis of popliteal reg ion is
found to be more positive for normal knee and negative or less
positive for early OA knee fro m image analysis.
The complete training and testing was performed rando mly
twice on the data collected. Figure 9 shows the confusion
matrix for two trials performed. Table 2 shows the average
statistical test data evaluation of diagnosis for the classified
output after two trials.
The authors wish to thank TEQIP Phase II for the grant
provided for taking up this research work. Authors also extend
thanks to Medical Electronics Depart ment of BMSCE, Dr.
Karthik Gajapathy and the staff of Sagar Clinic, Bangalore,
India for their support during the data collection.
REFERENCE
[1] RK Arya, Vijay Jain, ‘Osteoarthritis of the knee joint: An
overview’, JIACM 2013; 14(2): 154-62
[2] Richard F. Loeser, Age-Related Changes in the
Musculoskeletal System and
the Development of
Osteoarthritis, Clin Geriatr Med. 2010 August ; 26(3): 371–
386. doi:10.1016/j.cger.2010.03.002.
[3] Constance R Chu, Ashley A Williams, Christian H Coyle
and Megan E Bo wers, ‘Early diagnosis to enable early
treatment of p re-osteoarthritis’, Arthrit is Research & Therapy
2012, 14:212 http://arthritis-research.com/content/14/3/212
[4] Braverman IM. The cutaneous microcirculation. J Investig
Dermatol Symp Proc. 2000 Dec;5(1):3–9.
[5] Carolin Hildebrandt Christian Raschner and Kurt Ammer,
An Overview of Recent Application of Medical Infrared
Thermography in Sports Medicine in Austria , Sensors 2010,
10, 4700-4715; doi:10.3390/s100504700
[6] Christophe L Herry and Monique Frize, Quantitative
assessment of pain-related thermal dysfunction through
clinical digital infrared thermal imaging, BioMedical
Engineering On Line 2004, 3:19 doi:10.1186/1475-925X-3-19
Figure 9: Confusion matrix. Top: Tri al 1. B ottom: Trail 2
Table 2 Statistical test data evaluation of diagnosis for the
classified output
Measure
Value Deri vations
True Positive
0.8542 TPR = TP / (TP + FN)
Rate(Sensitiv ity)
True Negative
0.8889 TNR = TN / (FP + TN)
Rate (Specificity)
Positive Predict ive 0.8572 PPV = TP / (TP + FP)
Value (Precision)
Negative
0.8750 NPV = TN / (TN + FN)
Predictive Value
False Positive
0.1111 FPR = FP / (FP + TN)
Rate
False Discovery
0.1429 FDR = FP / (FP + TP)
Rate
False Negative
0.1459 FNR = FN / (FN + TP)
Rate
Accuracy
0.8665 ACC = (TP + TN) / (P + N)
F1 Score
0.8513
F1 = 2TP / (2TP + FP + FN)
International Journal of Engineering Science and Computing, August 2016
[7] Junghoon Lee, Joosung Lee, Sangha Song, Hyunsook Lee,
Kyoungjoung Lee, Youngro Yoon, Detection of Suspicious
Pain Regions on a Digital Infrared Thermal Image using the
Multimodal Function Optimization, 30th Annual International
IEEE EM BS Conference Vancouver, British Colu mb ia,
Canada, August 20-24, 2008
[8] Steven J Spalding, C Kent Kwoh, Robert Boudreau,
Joseph Enama, Julie Lunich,,Daniel Huber, Louis Denes and
Raphael Hirsch Three-dimensional and thermal surface
imaging produces reliable measures of joint shape and
temperature: a potential tool for quantifying arthritis,
Arthritis
Research
&
Therapy
2008,
10:R10
(doi:10.1186/ar2360)
[9] Sudhir Rathore and S. V. Bhalerao, Implementation of
Neuro-Fuzzy Based Portable Thermographic system for
Detection of Rheumatoid Arthritis, Proceedings of 2015
Global Conference on Communication Technologies(GCCT
2015)
[10]
Snekhalatha.U,
Anburajan.M,
Therace
Teena,
Venkatraman .B, Menaka.M, Baldev Raj, Thermal Image
2219
http://ijesc.org/
Analysis and Segmentation of Hand in Evaluation of
Rheumatoid Arthritis, 2012 International Conference on
Co mputer Co mmunication and Info rmatics (ICCCI -2012),
Jan. 10 – 12, 2012, Co imbatore, INDIA
[11] Nikola Borojevi, Darko Kolari, Simeon Grazio, Frane
Grubišic, Svetlana Antonini, Iskra Alexandra No la, Željko
Herceg, Thermography of Rheumatoid Arthritis and
Osteoarthritis, 53rd International Sy mposium ELMA R-2011,
14-16 September 2011, Zadar, Croatia
[12] Varju, G., Pieper, C.F., Renner, J.B. and Kraus, V.B.,
Assessment of hand osteoarthritis: correlation between
thermographic and radiographic methods, Rheumatology,
Vo l. 43 (2004), pp. 915-924.
[13] Ahlem Arfaoui, Mohamed Amine Bouzid, Hervé Pron,
Redha Taiar and Guillau me Po lidori, Application of Infrared
Thermography as a Diagnostic Tool of Knee Osteoarthritis,
Journal of Thermal Science and Technology .01/2012;
7(1):227-235. DOI: 10.1299/ jtst.7.227
[14] Denoble A E, Hall N, Pieper C F and Kraus V B 2010
Patellar skin surface temperature by thermography reflects
knee osteoarthritis severity Clin. Med. Insights: Arthritis and
Musculoskeletal Disord.
[15] Chao Jin, Yang Yang, Zu-Jun Xue, Ke -Min Liu and Jing
Liu, Automated Analysis Method for Screening Knee
Osteoarthritisusing Medical Infrared Thermography, Journal
of Medical and Biological Engineering(2012), 33(5): 471 -477
[16] Brian M. Sanchez, Mark Lesch, David Brammer, Susan
E. Bove, Melissa Thiel, Kenneth S. Kilgore, Use of a portable
thermal imaging unit as a rapid, quantitative method of
evaluating inflammation and experimental arthritis, Journal of
Pharmacological and To xico logical Methods 57 (2008) 169–
175
[17] Kellgren J, Lawrence J. Radiological assessment of
osteoarthrosis. Ann Rheum Dis 1957; 16:494–502.
[18] S. Theodoridis and K. Koutroumbas, “Nonlinear
classifier,”in: S. Theodoridis and K. Koutroumbas, (Ed.),
Pattern recognition. Canada: AcademicPress, 198-203, 2009.
International Journal of Engineering Science and Computing, August 2016
2220
http://ijesc.org/