Assessment of Disease Activity by Infrared Thermography in Patients with Rheumatoid Arthritis: A Comparative Cross-Sectional Study of Thermographic, Clinical and Ultrasound Assessments


 OBJECTIVES: To compare RA activity evaluation by thermal imaging with a camera adjustable to cellphones, standard clinical examination and ultrasound.MATERIALS AND METHODS: Monocentric study with 3 independent evaluations of RA activity: 1) tender (TJC) and swollen joints count (SJC) of wrists, MCP and PIP; 2) Ultrasound (US) examination with mode B and Power Doppler (PDUS); 3) thermographic assessment using infrared thermal cellphone camera FLIR One®. Thermal images analysis with (a) software detecting 10 regions of interest (ROI) corresponding to the studied joints and (b) by color reading (arthritis yes/no). For each ROI, temperature difference measurement “ΔT-joint” (ΔTj) compared to the ipsilateral forearm (reference area) and comparison of ΔTj to a) TJC and SJC; b) US synovitis in B mode or c) PDUS grade (0,1,2,3); d) synovitis in visual detection on thermographic image. T-tests and ANOVA were performed.RESULTS: 53 patients (43 women) with a mean (SD) age of 61.6 (12.3) years, totaling 921 examined joints. Mean disease duration was 17.6 (11.1) years. RA status was 81% ACPA+, 58% erosive, mean DAS28 score was 4.0 (1.6).Mean ΔTj was higher in tender (p<0.001), swollen (p=0.066) joints and in US synovitis in B mode (p=0.021). Mean ΔTj was not associated with PDUS category (p-ANOVA=0.072). Synovitis detected with thermal image reading was associated with PDUS grade 3 (p<0.001). CONCLUSION: Our results did not show the ability of thermal imaging to assess RA activity in small joints. However, our results are encouraging as temperature variation were observed in inflammatory joints.


Introduction
Regular evaluation of the activity of rheumatoid arthritis (RA) allows a strict control of the disease and treatment adaptation, to prevent joint destruction, impairment of quality of life and reduction of life expectancy (1)(2)(3)(4). RA activity can be evaluated by composite scores such as the Disease Activity Score 28 (DAS28), or imaging tools including ultrasound (US) or magnetic resonance imaging (MRI) (5)(6)(7).
Numerous composite scores include assessing the number of joints with active synovitis. However, evaluation of tender joint count (TJC) and swollen joint count(SJC) is poorly reproductible (8)(9)(10).
Osteoarticular US allows a good anatomical analysis in B mode and a vascular analysis in power Doppler US (PDUS) (11). Injected MRI is a reliable and sensitive tool to assess the in ammatory activity in RA (12)(13)(14)(15). However, these tools have limitations for frequent use, mainly for reasons of availability and cost for MRI. Ultrasound is less expensive and more easily available than MRI but requires time and a quali ed practitioner in osteoarticular US, with low intra and inter-observer reproducibility (16).
Thus, there is a need for an objective, rapid and low-cost measurement for an ambulatory evaluation of RA activity.
In the 1980s, several studies have evaluated infrared thermography and its capacity of assessing arthritis in patients with RA, mainly on large joints, through cutaneous temperature variation (17)(18)(19)(20). The increase of cutaneous temperature re ects the joint's temperature underneath. Thermography is a diagnostic imaging procedure that detects, records and produces a thermal image of the skin's surface temperature of the patient (21). Thermography does not involve ionizing radiation, venous access nor any other invasive procedure (21). There are currently two recognized methods of clinical thermographic imaging: electronic infrared thermography and liquid crystal thermography. The infrared thermography has the advantage of detecting contactless temperatures.
Mathematical formulas have been developed to represent temperature distribution on a thermal image. In 1982, Salisbury et al. used the Heat Distribution Index (HDI) from temperature's results of the thermal image to identify thermal synovitis (18). The HDI corresponded to ±1 standard deviation (SD) of the heat distribution following a Gaussian distribution. In 2008, Spalding et al. measured the temperature of two ROI in patients' hands and compared them to the measurement of joint volume (22). Even though they have shown the usefulness of thermography in identifying in amed joints, these thermal imaging cameras were large size and xed devices, thus limiting their use in current practice.
Recently, miniaturized thermal cameras have been developed, with technical capabilities similar to those of the previously cited studies. These devices can measure the temperatures ranging from 10 °C to 40 °C with a sensitivity of 0.1 °C. These miniature thermal cameras, adaptable to smartphones, have been largely marketed with no medical purpose but to detect leaks in buildings (23).
Some studies have already assessed RA activity, in particular in small joints, by infrared thermography through the use of miniaturized thermal cameras, but with discordant results (22,24). No study has compared thermography to ultrasound to date.
The aim of our study was therefore to compare the evaluation of RA activity by thermal imaging through miniaturized thermal cameras with clinical and ultrasound evaluation.

Design of study
We carried out a monocentric cross-sectional study from January to October 2019 in the Rheumatology Department of the University Hospital of Sainte Marguerite in Marseille, France. Inclusion criteria were patients suffering from RA (2010 ACR/EULAR criteria (25)), aged from 18 to 90 years old. The exclusion criteria were (a) the presence of Raynaud's syndrome, (b) a severe deformity of hands/wrist, (c) a recent surgery of wrist/ ngers within the 3 months preceding their inclusion, or (d) fever during clinical examination.

RA Activity Evaluation
Upon their arrival in the Rheumatology Department, each patient underwent an interview. The interview lasted about 10 minutes, in a standardized room, thus allowing the hands to rest. Then, the patient had 3 independent evaluations: rst, a thermal image was taken with the FLIR One® camera, then a clinical examination was performed by a second evaluator, and nally an ultrasound evaluation was done by a third evaluator. Each evaluator was unaware of the other assessments.
Thermal images were taken before physical examination and ultrasound evaluation to avoid measurement errors due to hand manipulation or the use of ultrasounds that could potentially modify the skin temperature.
In total, each patient had 3 independent assessments of joint disease activity of each hand (wrists, metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joints excluding thumbs): thermal evaluation, physical and US examination.

Clinical Evaluation
The following data were collected during the interview: patient age, body weight and height, medical history, current and previous treatments. RA history data were collected: disease duration, date of diagnosis, serological (rheumatoid factor (RF) and anti-CCP) and erosive status. The patient had then a full physical examination.
The examiner evaluated for each joint if they were painful (yes(Y)/ no(N)) and swollen (Y/N). The activity of the disease was evaluated with the overall Visual Analogue Scale (VAS), Tender Joints Counts (TJC), Swollen Joints Counts (SJC) and biological results (ESR (mm) and CRP (mg/L)).

Thermal evaluation with FLIR One® camera
An independent evaluator (DV) realized one thermal image per patient that included both hands.
Thermal images were analyzed in two different ways: (1) an automatic detection of the temperature in 10 ROI, through the software Gips Vision, and (2) a direct visual analysis of the thermal image.

Condition of acquisition of thermal images
Thermal images were obtained with a FLIR One® infrared camera adjustable to smartphones, FLIR T300 model. This camera has a 0.1 °C sensitivity and is originally used to detect leaks in buildings (23). Several settings of the camera had to be selected before taking thermal images, and we standardized the picture conditions in our routine (Supplementary Data 1).

Analysis of thermal images
Thermal images were analyzed in two different ways: (1) an automatic detection of temperature of 10 predetermined ROI, and (2) an analysis by direct visual detection using color reading of the thermal image (arthritis Y/N). In the absence of standard of cutaneous temperature, it was impossible to de ne a normal range of joint temperature for our measurements. Therefore, for each ROI we calculated the temperature difference (ΔTj) compared to the ipsilateral forearm, that was used as a normal reference temperature of the body (ΔTj = Temperature of joint -Temperature of ipsilateral forearm). Thus, the patient was considered as its own reference, and for each patient, we calculated the median of ΔT of all their joints.

Direct Visual Reading of the Thermal Image
After the automatic detection of ROIs temperature with the GipsVision® software, a second evaluation was made through a direct visual reading of the thermal image (Fig. 1). Indeed, this time the evaluator had to assess whether, in his opinion, there was an arthritis for each joint of interest by visual reading of the image. The thermal image was a color image in a scale from dark blue (for colder temperatures) to light yellow (for higher temperatures). Thus, arthritis was spotted in lighter yellow in comparison with other joints of the same and contralateral hand.

Statistical Analysis
As described before, in the absence of standard of cutaneous temperature, we took the patient as its own reference: for each ROI we calculated the temperature difference ΔT-joint (ΔTj) between each joint and ipsilateral forearm, the latter being considered as representative of the body temperature, and for each patient, we calculated the median of ΔT of all their joints, thereafter named "ΔT-patient" (ΔTp). Therefore, we calculated means and standard deviation (SD) of ΔTp at the patient level, and of ΔTj at the joint level.
At the joint level, we assessed the evaluation performances of clinical, ultrasound, and thermographic measures: we compared ΔTj, acquired from automatic detection of ROI using the GipsVision® software, according to a) presence or absence of pain or swelling; b) US synovitis in B mode or c) in PDUS mode (0,1,2,3); and d) synovitis in direct visual detection on thermographic image. T-tests for binary variables and ANOVA for characteristics with more than 2 categories were performed.
We also conducted a secondary analysis at the patient level. Without reference normal values of ΔT, we compared ΔTp across categories of DAS28 status (< 2.6; [2.6-5.1]; >5.1) by 1-way ANOVA and the CRP status (CRP < 5 or ≥ 5 mg/L) with a Student t-test.
All computations were performed using IBM® SPSS® version 20. All statistical tests were 2-sided and p < 0.05 was considered signi cant.

Ethical Approval Information
This study did not involve invasive evaluation on human participants as RA patients were included when coming to the Rheumatology Department for RA diagnosis or follow-up. There were no additional exams than those usually done as part of their regular follow-up. Therefore, this study did not justify the approval of Ethics Committee.
All participants received information and gave informed consent.

Results
Population characteristics 53 patients (43 women, 10 men) were included in our study, totaling 954 joints. Thirty-three joints, from 4 patients, had missing temperature data due to technical detection problem of ROI from the GipsVision® software. Therefore, a total of 921 joints were included in the analysis. The main population characteristics are presented in Table 1. Mean age (SD) was 61.6 (12.3) years, and mean disease activity of RA was moderate (mean DAS28 = 4.0), extending from 1.1 to 7.3.

Clinical examination vs. Automatic ROI thermal imaging
We compared clinical assessments of joint (presence or absence of pain or swelling) to the ΔTj acquired from automatic detection of ROI in thermal image. Of the 921 included joints, 268 were tender and 173 were swollen according to physical examination. The mean of ΔTj was signi cantly lower in non-tender joints than in tender joints (ΔTj= -0.30 °C ± 0.84 vs -0.06 °C ± 0.70, p < 0.001) and the same trend was observed, not at a signi cant level, between non-swollen and swollen joints (p = 0.066) (

Discussion
Our study compared the evaluation of RA activity by thermal imaging, through miniaturized thermal cameras, with clinical and ultrasound evaluation. Using an automatic detection of the ROI temperature, thermal imaging was correlated to tender joint count and US mode B results, but was not able to discriminate RA activity at the joint level for more relevant parameters of in ammation markers, i.e. swollen joints and PDUS.
Using direct visual identi cation of synovitis, thermal imaging showed only limited concordance with US measurements (mode B and PDUS).
There are few studies comparing clinical evaluation with thermal imaging from miniaturized thermal cameras, and their results are discordant.
In 2018, Jones et al. (24) compared thermal imaging of the hand joints of patients with RA to those of healthy volunteers. Their patients' characteristics were similar to ours. Thermographic analysis of joint temperature was not associated with clinical measures of disease activity. The calibration of their camera, the computer software, and the temperature assessment were not speci ed (24). The authors suggest that the clinical examination may lack sensitivity and that it would be interesting to compare it with US assessment.
Tan et al. (27) recently compared thermal imaging with US and clinical evaluation. Their sample was smaller (n = 37) with patients whose RA was more recent, but with the same activity as our population. They reported a signi cant association between thermal imaging and US data, but not with clinical examination. However, they performed assessment on all MCP and PIP joints, including thumbs. In our study, we made the choice to exclude thumbs as we wanted to avoid the bias of spotting thermal arthritis that could be either attributed to RA are or to trapezo-metacarpal osteoarthritis which is common on this joint. Neither the thermal camera used in Tan's study, nor the way of spotting joints and measuring their temperature, were described thus preventing detailed comparison with our study.
Despite the validity of our results, our study had several limitations. First, there was a lack of statistical power. Although including a large number of patients, the number of joints with the highest markers of RA activity was small compared to the number of joints with low activity (173 swollen joints vs 748 nonswollen, and 42 PDUS grade 3 joints vs 782 PDUS grade 0 joints). To increase the power of the study, it would require a greater number of active joints (swollen joints and PDUS grade 2-3 joints).
Another concern was the potential bias in PDUS evaluation. Five evaluators performed the ultrasound evaluation as our work was a daily practice observational study. We did not assess the intra and interreproducibility of the ultrasound evaluation.
The sensitivity of the thermal camera was also a potential limit. Indeed, the model used had a sensitivity of 0.1 °C (23). More recent cameras have been marketed since the start of our study, with a ner sensitivity (< 0.06 °C) (23). A new study with the use of a more sensitive thermal camera would be needed to better assess thermal arthritis, particularly those clinically swollen or presenting PDUS grade 2-3.
The methods used in the different studies to assess the presence of thermal synovitis from temperature data were heterogeneous, as HDI was used for Salisbury (18), ΔT in our study and Tan's (27), or not mentioned for Jones (24). A consensus on standardized measures would be required in future works.
Although our results did not show signi cant results on in ammation markers used for RA activity's evaluation (swollen-joint and PDUS-3), we observed a numerical trend for these parameters. Indeed, the temperature difference of joints with PDUS-3 was higher, meaning clinically "hotter", than those without doppler signal (PDUS-0). This trend was also found between the group with doppler signal (PDUS 2-3) and those without PD (PDUS 0-1). This was also the case for clinically swollen joints versus non-swollen joints.
Finally, thermal images were analyzed through automatic and direct visual detection. The results of both analyses were consistent, indicating that direct visual reading could be used without needing the software. This strengthens the thermal camera as a quick and easy tool to be used in ambulatory practice.

Conclusion
Our results did not show the ability of thermal imaging to assess RA activity in small joints. However, our results are encouraging with temperature variation observed in in ammatory joints. results interpretation and manuscript approval. VP: was in charge of the statistical analyses. All authors take responsibility for the integrity of the work as a whole, from inception to published article, and they should indicate that they had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. They give permission to reproduce published material, report sensitive personal information, to use illustrations of identi able persons or to name persons for their contributions.
Funding This study was supported by the CRI (Club des Rhumatismes et In ammation) The authors have not declared a speci c grant for this research from any funding agency in the public or commercial sectors.
Competing interests None declared.
Patient and Public Involvement was not appropriate in our study.
Ethical approval information: This study did not involve invasive evaluation on human participants as RA patients were included when coming to the Department of the University Hospital of Sainte Marguerite (Marseille, France) for RA diagnosis or follow-up. There were no additional exams than those usually done as part of their regular follow-up. Therefore, this study did not justify the approval of Ethics Committee. All participants received information on the study protocol and gave informed consent.
Data sharingstatement All data (clinical, US and thermal imaging) concerning RA patients and data analysis are available upon request in the Department of the University Hospital of Sainte Marguerite (Marseille, France).  Figure 1 Analysis for each thermal image: automatic detection and direct visual reading

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryMaterialall.docx