Detection of body temperature with infrared thermography: accuracy in detection of fever

Full text

Hong Kong Med J Vol 18 No 4 Supplement 3 August 2012 31
RESEARCH FUND FOR THE CONTROL OF INFECTIOUS DISEASES
Detection of body temperature with
infrared thermography: accuracy in
detection of fever
BMY Cheung 張文勇
LS Chan 陳龍生
IJ Lauder
CR Kumana 顧崇仁
Key Messages
1. Infrared thermography (IRT)
for detecting body temperature
is less accurate in women,
elderly people, and those with
fever.
2. The core temperature
signicantly but weakly
correlates to the IRT
temperatures obtained from
frontal and lateral of the face,
and the forehead.
3. Among the three areas, the
forehead IRT temperature
showed the largest discrepancy
and poorest correlation with the
core temperature.
4. If IRT is used, the lateral
maximum temperature of the
face should be used. A cut-off
temperature of 36ºC gives 77%
sensitivity and 74% specicity.
5. Owing to its weak correlation
with the core temperature, IRT
should not replace direct body
temperature measurement in
clinical situations.
The University of Hong Kong:
Department of Medicine
BMY Cheung, CR Kumana
Department of Earth Sciences
LS Chan
Department of Statistics and Actuarial
Science
IJ Lauder
RFCID project number: 03040232
Principal Investigator and corresponding author:
Prof Bernard MY Cheung
Department of Medicine, Queen Mary
Hospital, Hong Kong SAR, China
Tel: (852) 2255 4347
Fax: (852) 2818 6474
Email: mycheung@hku.hk
HongKongMedJ2012;18(Suppl3):S31-4
Introduction
Since the outbreak of severe acute respiratory syndrome, infrared thermography
(IRT) systems have been deployed at the airport and border crossings in Hong
Kong for screening travellers. However, its use to identify people with elevated
body temperature is limited. In a pilot study of 176 subjects,1 temperatures
measured by IRT might be used as a proxy for core temperature, but they are
affected by a variety of factors, such as the part of the face measured. We aimed
to investigate the effectiveness of IRT to identify people with fever.
Methods
This study was conducted from September 2005 to August 2006. The protocol
was approved by the Institutional Review Board of the Hong Kong West
Cluster of hospitals. Unselected patients attending the accident and emergency
department of the Queen Mary Hospital were invited to participate. Patients on
stretchers or needing immediate emergency treatment were excluded. Verbal
informed consent was obtained from each subject.
The core temperature was dened as either the oral or aural temperature,
or whichever was higher if both were available. At ports and border crossings,
the maximum IRT temperatures obtained from the frontal (Areamax) or lateral
(Latmax) of the face or the forehead temperature were used as proxies for the
core temperature. Ambient temperature, barometric pressure, and humidity were
also recorded. The degree of clothing and the time of measurement were noted.
For the study of the effect of distance on IRT readings, temperatures of 31
healthy (afebrile) volunteers were measured in a controlled laboratory setting
with the subjects standing at 1, 2, 3, 4 and 5 m from the IRT camera.
The software program ThermaCAM Researcher was used to extract from
the IRT temperatures of designated parts of the face. Data analysis was stratied
by age and gender. Pearson correlation coefcients between IRT temperatures
and oral/tympanic temperature were determined. The 95% condence limits
of agreement of IRT measurements with the reference method were calculated
according to the method of Bland and Altman.2 The standard error of the 95%
limit of agreement is approximately √(3s2/n), where s is the standard deviation of
the differences between measurements by the two methods, and n is the sample
size.2 The receiver operator characteristics were determined by plotting the
sensitivity against 1-specicity. The sensitivity, specicity, false-positive, and
false-negative rates of IRT were calculated. Likelihood ratios, which describe
the odds of getting a positive or negative test result, were calculated from the
sensitivity and specicity.
Results
A total of 747 men and 770 women consented to participate; 215 of them had
a core temperature of ≥37.5ºC and were considered to have fever. The forehead
IRT temperature showed the largest discrepancy from the core temperature and
was on average 3.1ºC lower. The Latmax yielded the best correlation with the

Cheung et al
32 Hong Kong Med J Vol 18 No 4 Supplement 3 August 2012
core temperature (r=0.441), whereas the forehead IRT
temperature yielded the poorest correlation (r=0.361)
[Table 1].
In all subgroups examined, forehead IRT temperature
was consistently lower than Latmax or Areamax (Fig 1). The
difference between core and IRT temperature was greatest
in febrile subjects; the forehead IRT temperature was on
average 3.0ºC and 3.7ºC lower than the core temperature in
afebrile and febrile subjects, respectively (Fig 1).
In the Bland-Altman plots of the difference between the
IRT and core temperatures against the mean of the IRT and
core temperature, IRT temperatures were on average lower
than the core temperature. The difference between IRT and
core temperatures increased as core temperature decreased
(Fig 2).
The subjects were divided into nine age groups (1-
2, 3-6, 7-10, 11-19, 20-29, 30-39, 40-49, 50-65, 66-100
years). The best correlation of IRT temperatures with
core temperature was seen in children (aged 3-18 years),
followed by infants (aged 1-2 years). Male subjects showed
better correlation between IRT and core temperatures. The
respective correlation coefcients for the three variables of
Areamax, Latmax, and Forehead were 0.496, 0.5, and 0.404
for males, and 0.369, 0.385, and 0.323 for females (Table
1). A better correlation was observed in subjects with a core
temperature of ≥37.5ºC. For subjects with a normal body
temperature, the correlation coefcients between the IRT
and core temperatures tended to be <0.25.
Ambient temperature had a minor effect on IRT values.
Each 1ºC change in ambient temperature changed the IRT
values by 0.196ºC on average.
The sensitivity, specicity, type-I error, and type-II error
at different IRT temperatures are tabulated in Table 2. At
36ºC, the positive and negative likelihood ratios were 3.97
and 0.39 for the Latmax, respectively.
Table 1. Mean infrared thermographic (IRT) temperatures for the frontal (Areamax) and lateral (Latmax) of the face and the
forehead, and correlation coefficients (r) between IRT and core temperatures
Parameter Areamax (n=1511) Latmax (n=1513) Forehead (n=1509)
Mean±SD IRT temperature (ºC) 35.23±0.99 35.43±1.03 33.79±1.15
Mean±SD difference from core temperature (ºC) -1.67±0.93 -1.46±0.96 -3.10±1.11
Mean±SE lower limit of agreement -3.49±0.04 -3.34±0.04 -5.28±0.04
Mean±SE upper limit of agreement 0.15±0.04 0.42±0.04 -0.92±0.04
r
for all 0.434 0.441 0.361
r
for males 0.496 0.500 0.404
r
for females 0.369 0.385 0.323
Fig 1. Mean and standard deviation of core and infrared thermography temperatures in different subgroups
40
39
38
37
36
35
34
33
32
31
30
All Male Female 1-2
years
old
3-6
years
old
7-10
years
old
11-19
years
old
20-29
years
old
30-39
years
old
40-49
years
old
50-65
years
old
66-100
years
old
Normal Normal-
young
Febrile-
young
Normal-
old
Febrile-
old
Febrile
Core temperature Areamax Latmax Forehead
Temperature (°C)
Group

Detection of body temperature with infrared thermography
Hong Kong Med J Vol 18 No 4 Supplement 3 August 2012 33
Distance between subject and IRT had a signicant
effect on IRT readings; IRT temperature decreased linearly
with distance (p=0.001). Using 1 m as the reference, the IRT
temperature was 0.35ºC lower at 2 m and 1.1ºC lower at 5
m. The IRT temperature decreased on average by 0.26ºC
per meter of distance.
Discussion
The correlation of IRT temperatures with the core
temperature was signicant but weak (r<0.45). Gender,
age, and core temperature inuenced the accuracy of IRT
temperature as a proxy for body temperature. Females
showed a poorer correlation between IRT and core
temperatures. It is not possible to rule out if this was due
to cosmetics, as such data were only available on three
subjects.
The IRT system seems more accurate in younger age
groups, especially children and teenagers.3-5 The core
temperatures were higher in children than adults, perhaps
because children with fever were more likely to attend
hospital. The core temperatures in the elderly were lower,
and their febrile response to infection could be attenuated.
The Bland-Altman analysis showed that IRT
temperatures were lower than the core temperature,
especially when the core temperature was low. This nding
may be useful as it reduces the number of people with a
normal core temperature being mistaken for having fever.
The use of forehead IRT temperature as a proxy for
the body temperature is questionable.6 The forehead IRT
temperature was lowest among the three IRT temperatures
of the face. Its correlation with the core temperature was
also lowest. Based on the forehead IRT readings, if 37ºC
was used as the cut-off temperature for screening, the
sensitivity was exceedingly low (4%). Reducing the cut-off
temperature to 36ºC and 35ºC increased the sensitivity to
25% and 52%, respectively. To achieve a sensitivity of about
79%, the cut-off temperature should be lowered to 34ºC.
This, however, would yield a specicity of 55% and a false
positive rate of 88% (88% of those tested positive would
actually be afebrile). This would require an unacceptably
high percentage (47.8%) of subjects to be retested. Thus,
the forehead IRT temperatures are not effective in screening
passengers with fever. This casts doubt on the efcacy of
using a single-point IRT probe to detect passengers with
fever.
33 34 35 36 37 38 39 40 41
Mean core & IRT temperatures (°C)
8
6
4
2
0
-2
-4
Core - IRT temperatures (°C)
Areamax
33 34 35 36 37 38 39 40 41
Mean core & IRT temperatures (°C)
8
6
4
2
0
-2
-4
Core - IRT temperatures (°C)
Latmax
33 34 35 36 37 38 39 40 41
Mean core & IRT temperatures (°C)
8
6
4
2
0
-2
-4
Core - IRT temperatures (°C)
Forehead
Table 2. Sensitivity and specificity of maximum frontal and lateral infrared thermographic (IRT) temperatures
Parameter Cut-off temperature
34ºC 34.5ºC 35ºC 35.5ºC 36ºC 36.5ºC 37ºC 37.5ºC
Maximum frontal IRT temperature
Sensitivity 0.97 0.94 0.88 0.79 0.68 0.52 0.40 0.21
Specificity 0.08 0.20 0.39 0.60 0.83 0.96 0.99 1.00
Type-II error, β0.03 0.06 0.12 0.21 0.32 0.48 0.60 0.79
Type-I error, α0.92 0.80 0.61 0.40 0.17 0.04 0.01 0.00
False negative rate 0.02 0.02 0.02 0.03 0.03 0.04 0.05 0.06
False positive rate 0.92 0.91 0.90 0.86 0.76 0.51 0.21 0.08
Failing %* 92.0 80.9 63.3 42.8 20.7 7.9 3.7 1.7
Maximum lateral IRT temperature
Sensitivity 0.97 0.96 0.89 0.85 0.77 0.61 0.46 0.21
Specificity 0.07 0.17 0.31 0.52 0.74 0.90 0.97 1.00
Type-II error, β0.03 0.04 0.11 0.15 0.23 0.39 0.54 0.79
Type-I error, α0.93 0.83 0.69 0.48 0.26 0.10 0.03 0.00
False negative rate 0.03 0.02 0.03 0.02 0.02 0.03 0.04 0.06
False positive rate 0.92 0.92 0.91 0.88 0.81 0.67 0.43 0.23
Failing %* 92.9 84.0 70.9 51.0 29.4 13.5 6.1 2.0
* Total percentage of subjects tested positive
Fig 2. Bland-Altman plots of the difference between core and infrared thermography (IRT) temperatures of the frontal (Areamax)
or lateral (Latmax) of the face or the forehead against the means of core and IRT temperatures

Cheung et al
34 Hong Kong Med J Vol 18 No 4 Supplement 3 August 2012
When the maximum frontal IRT temperature was used
as the screening temperature, a cut-off temperature of 36ºC
would yield a sensitivity of 68% and would result in 22.4%
of all subjects to fail the screening. This is much better than
the forehead IRT temperature in terms of sensitivity and
retesting rate. Reducing the cut-off temperature to 35.5ºC
would yield a sensitivity of 79% and a specicity of 60%.
However, 86% of those tested positive would actually be
afebrile and the percentage of subjects failing the screening
would increase to 51%.
When the maximum lateral IRT temperature was used
as the screening temperature, the same cut-off temperature
of 36ºC would yield a sensitivity of 77%, a specicity of
74%, and a false negative rate of 23%. This would be a
reasonable setting in terms of sensitivity and false negative
rate. However, it would require 29.4% of the subjects to
be retested. If the percentage of subjects requiring retesting
is a constraining factor, raising the cut-off temperature to
36.5ºC would reduce the percentage of subjects failing
the screening to 13.5%. However, the sensitivity would
be reduced to 61% and the false negative rate increased to
39%. This may be unacceptable during an epidemic.
The distance between the IRT camera and the subject
is a limiting factor on the efciency. Although the camera
can be calibrated for different distances, it is impractical
at border crossings and airports to do so. One particular
mode of operation compares the maximum detected
temperature of travellers passing in front of the camera
with the temperature inside a control box kept at a constant
temperature. Errors can arise if the subject and the control
box are at different distances from the camera.
Conclusions
For the application of IRT in screening for travellers with
elevated body temperature at airports and border crossings,
the forehead IRT temperature differed substantially from the
core temperature, and the maximum lateral IRT temperature
should be used. The reading should also be taken at a
dened distance from the camera. Overall the sensitivity of
IRT in detecting fever is low unless the cut-off temperature
is low. When the risk of an epidemic is high and high
sensitivity is required, a low cut-off temperature (≤35.5ºC)
should be chosen, although a large number of people will
require a conrmatory temperature measurement. As IRT is
relatively less accurate on women and older people, more
sampling for aural measurement should be done on these
individuals.
Acknowledgements
This study was supported by the Research Fund for the
Control of Infectious Diseases, Food and Health Bureau,
Hong Kong SAR Government (#03040232). We thank Ms
Jessica Lo (research nurse) and Ms Maggie Chan (research
assistant/data analysis).
References
1. Chan LS, Cheung GT, Lauder IJ, Kumana CR, Lauder IJ. Screening
for fever by remote-sensing infrared thermographic camera. J Travel
Med 2004;11:273-9.
2. Bland JM, Altman DG. Statistical methods for assessing agreement
between two methods of clinical measurement. Lancet 1986;1:307-
10.
3. Falzon A, Grech V, Caruana B, Magro A, Attard-Montalto S.
How reliable is axillary temperature measurement? Acta Paediatr
2003;92:309-13.
4. Craig JV, Lancaster GA, Taylor S, Williamson PR, Smyth RL.
Infrared ear thermometry compared with rectal thermometry in
children: a systematic review. Lancet 2002;360:603-9.
5. Ng DK, Chan CH, Lee RS, Leung LC. Non-contact infrared
thermometry temperature measurement for screening fever in
children. Ann Trop Paediatr 2005;25:267-75.
6. Liu CC, Chang RE, Chang WC. Limitations of forehead infrared
body temperature detection for fever screening for severe acute
respiratory syndrome. Infect Control Hosp Epidemiol 2004;25:1109-
11.