See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/323159298

Low cost thermal camera for use in preclinical detection of diabetic peripheral
neuropathy in primary care setting

Conference Paper · February 2018

DOI: 10.1117/12.2290335

CITATIONS READS

0 142

10 authors, including:

Zyden Jarry Mark Burge

visionquest biomedical University of New Mexico
4 PUBLICATIONS   11 CITATIONS    229 PUBLICATIONS   5,486 CITATIONS   

SEE PROFILE SEE PROFILE

Peter Soliz Vinayak S. Joshi

VisionQuest Biomedical VisionQuest Biomedical LLC
85 PUBLICATIONS   1,190 CITATIONS    26 PUBLICATIONS   206 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hopspital Databse View project

Umbra development View project

All content following this page was uploaded by Vinayak S. Joshi on 06 September 2018.

The user has requested enhancement of the downloaded file.

 Low cost thermal camera for use in preclinical detection of diabetic
peripheral neuropathy in primary care setting
V. Joshi*a, N. Manivannana, Z. Jarrya, J. Carmichaela, M. Vahtela, G. Zamoraa, C. Calderb,
J. Simonc, M. Burged, P. Soliza
a
VisionQuest Biomedical LLC., Albuquerque, NM, USA 87106; bDepartment of Neurology,
University of New Mexico, Albuquerque, NM, USA 87131, cFoot and Ankle Associates of New
Mexico, Albuquerque, NM, USA 87122, dDepartment of Internal medicine, University of New
Mexico, Albuquerque, NM, USA 87131

ABSTRACT

Diabetic peripheral neuropathy (DPN) accounts for around 73,000 lower-limb amputations annually in the US on
patients with diabetes. Early detection of DPN is critical. Current clinical methods for diagnosing DPN are subjective
and effective only at later stages. Until recently, thermal cameras used for medical imaging have been expensive and
hence prohibitive to be installed in primary care setting. The objective of this study is to compare results from a low-
cost thermal camera with a high-end thermal camera used in screening for DPN.

Thermal imaging has demonstrated changes in microvascular function that correlates with nerve function affected by
DPN. The limitations for using low-cost cameras for DPN imaging are: less resolution (active pixels), frame rate,
thermal sensitivity etc. We integrated two FLIR Lepton® (80x60 active pixels, 50° HFOV, thermal sensitivity <
50mK) as one unit. Right and left cameras record the videos of right and left foot respectively. A compactible
embedded system (raspberry pi3 model Bv1.2) is used to configure the sensors, capture and stream the video via
ethernet. The resulting video has 160x120 active pixels (8 frames/second).

We compared the temperature measurement of feet obtained using low-cost camera against the gold standard high-
end FLIR® SC305. Twelve subjects (aged 35-76) were recruited. Difference in the temperature measurements between
cameras was calculated for each subject and the results show that the difference between the temperature measurements
of two cameras (mean difference=0.4, p-value=0.2) is not statistically significant. We conclude that the low-cost
thermal camera system shows potential for use in detecting early-signs of DPN in under-served and rural clinics.

Keywords: Diabetic Peripheral Neuropathy, Diabetes, Thermal Imaging, Low-cost Screening, Primary care clinic

1. INTRODUCTION
Diabetes affects over 400 million people worldwide [1] and 3 million in the US. A comorbidity of diabetes is diabetic
peripheral neuropathy (DPN) [2]. Recently, the US and other countries have made progress in the management of
complications from diabetes. Vision loss from diabetes can be prevented or delayed through periodic eye screening.
Today over 50% of the US population with diabetes receive an annual eye examination or are screened for signs of
diabetic retinopathy. However, today tests to assess DPN in the primary care clinic do not provide a repeatable, non-
invasive means for detecting DPN in the preclinical stage. Tests, such as nerve biopsies [3] [4] can reveal evidence of
DPN, but are invasive and not appropriate for repeated clinical use in primary care settings. Nerve conduction velocity
(NCV) tests [5] [6] have been used as alternative, but such tests are not available in most primary care clinics. Sensation
tests that are commonly utilized in primary care clinic, such as vibration perception and monofilament test for pressure
sensation for DPN, have been shown to lack sensitivity in the early stages of DPN [4].

An alternate approach for identifying preclinical DPN is based on the characterization of thermoregulation as
expressed by changes in blood flow in the plantar foot. The mechanistic process is thermoregulation of homeostasis.
Exposure to cold stimulus causes the skin blood flow to decrease via cutaneous vasoconstriction. Removal of the cold
stimulus results in the eventual return of pre-stimulus rates of blood flow. However, for this technique to be adopted
by the primary care clinics, the procedure must be simple and the device to be low-cost. This study demonstrates not
only a highly sensitive and specific procedure, but implements a low-cost thermal device that is affordable by most
clinics and screening sites.
Previous research studies have found that there is a link between changes in the microvascular circulation in the
presence of peripheral neuropathy and changes in the functioning of the thermoregulation system (temperature
recovery characteristics) [7]. In this research, modern high speed thermal imaging technology was used for its ability
to record the most critical temperature changes which occur immediately and rapidly (exponentially) after removal of
the cold provocation. No investigator has previously presented a robust method as a means for the classification of
DPN in the plantar foot thermal data. Most measurements to date have failed to capture the necessary and appropriate
data to fully characterize the exponential recovery nature of the thermal auto-regulation for the plantar foot [8] [9].
Most previous studies have tried to discover classification patterns based on a single (static) or a few images in time
with or without a cold provocation or in temporal data acquired at long intervals (minutes) [9]. Because of the varying
nature of how DPN presents in different patients and at different early stages of the disease, no single approach will
capture all these variations.

Previous research has applied IR thermosensors, IR camera systems, and shoes that measure temperature, and liquid
crystal thermography for assessing the diabetic foot [10]. The research shows that these systems indicate that
temperature is an important factor, and is a potential monitoring system for ulceration in a diabetic foot; but have not
looked at the neuropathy indications, or the comparisons between types of thermals technologies [10]. In order to
provide a reliable diagnosis and staging for DPN in primary care settings, it is also essential to consider the critical
characteristics of the thermal imaging process, as well as the usability features of the device such as affordability and
accessibility. A low-cost imaging device with high sensitivity, high specificity and a well-defined measurement
protocol will significantly contribute to evaluation and treatment of diabetic foot.

2. METHOD
2.1 Equipment setup
The experiment utilized two thermal imaging cameras. The first one, which was considered as a standard for thermal
imaging, was a high-end thermal camera, FLIR Model SC 305. This camera works within infrared wavelength of 7.5
to 13 µm and was calibrated to give a thermal sensitivity of < 0.05 ○C at an imaging frequency of 9 Hz. The temperature
changes were measured in the order of 5-10 ○C; thus, measurement errors were in the order of 0.1%. The FLIR camera
was adopted in this study as the standard for thermal imaging due to its superior imaging capabilities compared to the
other camera. The second camera, “Lepton”, was sold by FLIR Systems, Ind.; that provides 50 degrees field of view,
an LWIR sensor with wavelength between 8 to 14 µm, effective frame rate of 8.7 Hz, and thermal sensitivity of <50
mK. Both the cameras were mounted on a portable platform, which allows for movement of the patient who sits in a
recliner (see Figure 1). The platform has adjustable foot braces to reduce movement of the foot.

Figure 1: Thermal imaging setup

2.2 Data
Thermal imaging of the plantar foot was performed, on 12 subjects (9 males and 3 females) between the ages of 35
and 76 years. A total of 5 subjects were diagnosed as DPN, 3 with diabetes mellitus (DM) only but no DPN diagnosis,
and 4 controls with neither DPN nor DM. DPN patients were categorized based on physician’s (MD) diagnoses prior
to entry in the study. The DPN diagnosis was typically done through a combination of clinical exams, including the
vibration threshold test, monofilament testing, and patient reported symptoms of pain or loss of sensation. Personal
health information was collected from each subject, including: age, sex, diabetes history and type; smoking and alcohol
consumption habits, blood pressure, body mass index, medications, recent foot surgery, exercise habits, foot care
habits, and history of any cardiovascular diseases (including peripheral vascular disease).

2.3 Thermal imaging protocol

All subjects rendered written informed consent as approved by the University of New Mexico Human research Review
Committee (Study # 15-494) and/or the Ethical and Independent Review Services Institutional Review Board (Study
# 15085-01A). The study participants were consented, and understanding of study activities was ensured. Following
consent, the participants were asked to enter the assessment room, which has a maintained temperature of 22 ± 2°C.
The participant’s blood pressure was taken, followed by measurements of height and weight. The participants were
asked to remove their socks and shoes, and then seated in a recliner with their feet elevated (see figure 1). The foot
mount was placed so the feet were between the braces, and camera had a centered image of the feet. The patient’s feet
temperature (°C) was taken, and the monofilament exam was performed. Following the monofilament exam, the
vibration threshold test was performed. Participants’ feet were then secured in the braces (Figure 1) with the feet
upright and flat as possible for patient comfort. The baseline 3-minute thermal imaging video was taken for each foot.
The cold provocation ice pack (coolant) temperature was measured, and both left and right foot packs were ensured to
be at 4.5°C, ± 2°C. The packs were secured to the participant’s feet using a pack holder with arch support to ensure
full foot coverage. Following the baseline video, the pack holder was secured in place using Velcro straps, for 5
minutes. The thermal camera was set, and immediately after removal of the ice packs (5 seconds), a 15-minute recovery
video was captured. The participants were asked questions regarding health history, medications, exercise, etc. and
then given time to rest or read until the video concluded.

Figure 2: Regions of
interest on the foot

For DPN diagnosis, all subjects received comprehensive clinical examination; which included visual inspection of
feet; ankle brachial index calculated from blood pressure readings at the ankle and arm; Semmes-Weinstein 10 gram
monofilament to test protective sensation on six points (see Figure 2) of the plantar and one point on the dorsum of
both feet; a 128 Hz tuning fork vibration sensation test on four bony prominences of both feet (i.e., the medial and
lateral malleoli of the ankles, and medial and lateral metatarsal joints); pin prick sensation and point temperature
measurement of the dorsum of each foot using an infrared thermometer. Sympathetic skin response tests were also
performed on all subjects. All medications being used were documented. Individuals who smoke or had uncontrolled
hypertension, had ongoing, unhealed foot ulcers or had amputations were excluded.

3. DATA ANALYSIS
3.1 Pre-processing of thermal data
Thermal videos obtained from both cameras need to be preprocessed to obtain meaningful thermal signal. At first, the
‘.seq’ files provided by the cameras were converted to ‘.mat’ files for further processing. A temperature calibration
matrix provided by the respective camera manufacturers were used to covert the thermal signals to the temperature
units; where the ‘.mat’ files recorded the temperature at each pixel in Celsius. The video frames were then analyzed to
determine the temperature before cooling (last frame of baseline video) and the temperature after cooling (initial frame
of 15-minute video). Figure 3 shows the baseline frame as well as temperature recovery frames for control and DPN
subjects. The six recovery points (figure 2) were taken from initial frame as a reference and rest of the video frames
were registered.

Figure 3: Temperature recovery patterns for

control (top) and DPN subject (bottom),
demonstrating higher variability in the recovery
pattern for DPN subject.

3.2 Temperature recovery measurements

The temperature recovery measurements were then obtained at six recovery points, at every 5 seconds. Figure 4
demonstrates the temperature recovery measurements captured at four regions on foot (Ball-2, internal/medial arch,
external/lateral arch, heel) from two cameras that show comparable temperature curves; although, the Lepton camera
demonstrates a step-increase of temperature occurring at specific time points. The average step increase in temperature
is 0.83° and it appears to occur every two minutes. It was found to be due to automatic gain correction (AGC) mode
in Lepton module. Furthermore, the data obtained from Lepton camera appears noisy compared to the one obtained
from FLIR. However, such a data can be preprocessed using advanced signal and image processing techniques to make
it equally useful as the FLIR camera data. Therefore, further experiments will be needed to develop the data
preprocessing methods for Lepton camera signals.

Figure 4: Temperature recovery measurements for four regions on left foot of the same patient, using: a) FLIR, b) Lepton
 4. RESULTS
4.1 Comparison of temperature recovery measurements
To compare the performance of two cameras in measuring temperature recovery characteristics (temperature changes
after cooling) using thermal videos, the temperature measurements were obtained at all of the six regions of interest
(ROIs) shown in figure 2 (ball-1, ball-2, ball-3, internal/medial arch, external/lateral arch, heel). The temperature
measurement after cooling (at initial frame) at all ROIs of both feet demonstrated no significant difference when
compared between two cameras (p > 0.05). Similarly, the percentage temperature recovery during the first minute after
cooling at all ROIs of both feet showed no significant difference as well (p > 0.05). For a visual comparison of
performance of two cameras, figure 5 shows the Bland Altman plot for temperature recovery measurements in one of
the ROIs, Ball-2 area, [1]. The Bland Altman plot provides a visual interpretation of an agreement between the
temperature measurements performed by two cameras. As demonstrated in figure 5, this plot shows the mean of two
temperature measurements performed by two cameras; against the difference (Δ) between those measurements. This
plot allows for assessment of agreement between two cameras as well as the distribution of error. Similarly, as
demonstrated in figure 6, the temperature recovery measurements performed by two cameras in Ball-2 area show a
near-to-perfect correlation of ϒ=0.98.

Figure 5: Bland Altman plot Figure 6: Correlation plot

In order to analyze the performance of two cameras (FLIR and Lepton) in individual disease groups, the temperature
recovery measurements were also compared in DPN, DM, and control groups, using Student’s T-test, as shown in
Table 1. The p-values for all the population groups demonstrated no statistically significant difference (p > 0.05).

Table 1: Temperature
recovery comparison

Population P-value
group
Control 0.20
DM 0.17
DPN 0.37

5. DISCUSSION
This study showed a quantitative comparison of performance of two thermal imaging cameras in measuring
temperature recovery characteristics of study subjects with DPN, DM, or controls. The correlation coefficient of
ϒ=0.98 between the performance of two cameras and a T-test with p-values of >0.05; demonstrate that there is no
significant difference between the temperatures measured by the two cameras. Essentially, a low-cost camera such
Lepton may perform comparably with high-end cameras such as FLIR, for the clinical application of detecting DPN.
The Bland–Altman plot (Figure 5) shows that both cameras have no significant differences in performance as the mean
difference between temperature measurements is close to 0. However, the plot shows three outliers (each with
approximate temperature difference of 2.7 degrees) without which the 95% limits of agreement would have been
“tighter”. On thorough investigation of these outlier subject’s data, it was found that the preprocessing of thermal
videos of these subjects may have introduced this error, which usually happens when subject move their feet during
the thermal imaging process. It shows that there is still room for improvement in the preprocessing techniques.
The Bland–Altman plot also shows the average difference in temperature measurements by two cameras is 0.73
degrees. This difference represents 3.4% of the average temperature values measured by Lepton camera; and we expect
that such a small difference in temperature measurement may not affect the overall diagnosis of DPN when using the
temperature recovery values.
Although, Lepton camera performs comparably with FLIR in measuring temperature recovery characteristics in all
groups of study subjects (DPN, DM, controls); being a low-cost camera, Lepton camera demonstrates limited
performance in terms of resolution, temperature sensitivity etc. However, we believe that with appropriate
preprocessing of thermal video data and advanced image processing techniques, the thermal signal captured by Lepton
camera should be adequate to determine the DPN status of subjects. Nonetheless, this will need a large dataset captured
by Lepton as well as FLIR cameras, to develop the image processing algorithms and compare the performance of DPN
detection by two cameras. Potentially, a better evaluation of performance of Lepton camera in detecting DPN would
be comparing it to clinical ground truth. We would like to do this in the near future.
The DPN is a clinically well-known comorbidity of diabetes mellitus. However, the clinical detection of DPN has been
based on subjective and highly flawed testing methods, which cannot provide quantitative and robust measure of
presence of DPN. Although, the thermal imaging modalities have been proven to provide more objective detection of
DPN, they are not affordable and easily accessible to a large number of primary care clinics, especially in rural or
remote regions. Our approach demonstrates a clinical utilization of a low-cost thermal camera that performs
comparably with a high-end thermal camera, in measuring the temperature characteristics for detection of DPN, and
it is more cost-effective for a primary care use. This approach has the potential to have a significant impact on the
early detection and treatment of DPN, making it available for mass screening.

6. CONCLUSIONS AND FUTURE RESEARCH

In summary, we have demonstrated the operation and performance of a low-cost thermal imaging camera “Lepton” in
measuring the temperature recovery characteristics for detection of DPN. The Lepton camera is capable of estimating
the temperature measurements with accuracy comparable to the high-end thermal cameras such as FLIR. In future, we
will test the performance of Lepton and compare it with other high-end cameras and the clinical ground truth, on larger
dataset.

ACKNOWLEDGMENT

This work was funded by NIDDK grant 2R44DK104578-02.

REFERENCES

[1] World Health Organization, "Global Report on Diabetes," World Health Organization,
Geneva, (2016).
[2] Center for Disease Control, "National Diabetes Statistics Report," Center for Disease
Control, Atlanta, (2017).
[3] Timar, B., "The usefulness of quantifying interapidermal derve fibers density in the
diagnostic of diabetic peripheral neuropathy: a cross-sectional study," Diabetology &
Metabolic Syndrome, vol. 8, no. 31, pp. 1-7, (2016).
 [4] Malik, R. A., "Which test for diagnosing Early Human Diabetic Neuropathy," Diabetes,
vol. 63, pp. 2206-2208, (2014).
[5] Lee, J.A., "Reliability and validity of a point-of-care sural nerve conduction device for
identification of diabetic neuropathy," PLOS One, vol. 9, no. 1, pp. 1-7, (2017).
[6] Mochizuki, Y., "Association of peripheral nerve conduction in diabetic neuropathy with
subclinical left ventricular systolic dysfunction," Cardiovascular Diabetology, vol. 14, no.
47, pp. 1-9, (2015).
[7] Kellogg, D.L., "In vivo mechanisms of cutaneous vasodilation and vasoconstriction in
humans during thermoregulatory challenges," Journal of Applied Physiology, vol. 100, pp.
1709-1718, (2006).
[8] Barriga, E.S., "Computational basis for risk stratification of peripheral neuropathy from
thermal imaging," Engineering in Medicine and Biology Society (EMBC), Annual
International Conference of the IEEE, (2012).
[9] Chekh, V., "Quantitative Early Detection of Diabetic Foot," ACM-BCB , pp. 86-95, (2013).
[10] Roback, K., "An overview of temperature monitoring devices for early detection of diabetic
foot disorders," Expert Reviews Medical Devices, vol. 7, no. 5, pp. 711-718, (2010).
[11] Johnson, J.M., "Local thermal control of the human cutaneous circulation," Journal of
Applied Physiology, vol. 109, no. 4, pp. 1229-1238, (2010).
[12] Smith, A.G., "The Diagnostic Utility of Sudoscan for Distal Symmetric Peripheral
neuropathy," Journal of Diabetes Complications, vol. 28, no. 4, pp. 511-516, (2014).

View publication stats