Innovations in Pediatric Rheumatology: Using Computer Vision for Infrared Detection of Joint

Inflammation

Abstract:

Pediatric rheumatic diseases often present diagnostic challenges due to their varied symptoms and the limitations

of traditional assessment methods. This study introduces an innovative approach using infrared thermography

combined with computer vision to enhance diagnostic accuracy in detecting pediatric joint inflammation. The

study included 18 pediatric patients with acute knee arthritis and 19 healthy controls. A smartphone-based FLIR

ONE Edge Pro thermal camera was employed for infrared thermography, and a custom Python script with

OpenCV was utilized for image analysis. The mean T max of the right knee in arthritis patients was significantly

higher compared to healthy children (33.37±1.57 vs 31.73±1.23, p = 0.001). Similar patterns were observed in

other temperature measures and the left knee. The color spectrum analysis revealed that median pixel values in

the red spectral band, associated with heat due to inflammation, were markedly higher in patients than in controls,

showing statistical significance (p < 0.05) and substantial effect sizes (r). The ROC analysis indicated high

diagnostic accuracy with AUC values of 0.822 (95% CI: 0.720-0.924) for knee T max and 0.797 (95% CI: 0.690-

0.904) for knee T ave. The integration of infrared thermography with computer vision demonstrates a promising,

non-invasive diagnostic tool for pediatric rheumatology. This method offers enhanced accuracy in detecting joint

inflammation, potentially transforming the management and treatment of children with rheumatic diseases.

1
Introduction

Pediatric rheumatic diseases, a diverse spectrum of musculoskeletal and systemic disorders, present with a pattern

of exacerbations and remissions, profoundly affecting children's quality of life. 1 Acute arthritis, characterized by

joint swelling, redness, and warmth, is a common inflammatory manifestation in this group. 2 Identifying the

cause of childhood inflammatory arthritis (IA), whether it be juvenile idiopathic arthritis (JIA), septic arthritis, or

others, requires a multi-faceted approach involving clinical examination, serologic and imaging studies. 2, 3

Traditional assessment methods, while foundational, have limitations—physical examinations may underestimate

inflammation, and serologic markers, though useful, do not always offer conclusive evidence. 4, 5

Imaging modalities have their own drawbacks; for instance, X-ray and MRI are encumbered by radiation

exposure and the use of contrast materials, respectively. Ultrasound, though adept at early detection of joint

anomalies, is limited by the lack of standardized pediatric references and operator variability. 4, 5 These challenges

underscore the critical need for timely and accurate diagnoses, which dictate the trajectory of treatment and

management.

In the field of pediatric rheumatology, innovation has paved the way for infrared thermography. This novel, non-

invasive method employs thermal cameras to measure cutaneous temperature changes emanating from inflamed

joints, identifying heat signature of conditions such as rheumatoid arthritis and systemic lupus erythematosus

(Figure 1).6, 7 The heat patterns captured by the camera provide a physiological heat-map, invaluable for detecting

temperature anomalies indicative of inflammation. 6, 7 Recent studies, such as those by Lasanen and colleagues,

have demonstrated the utility of thermal imaging in detecting joint inflammation in patients with conditions like

JIA or autoimmune diseases, including systemic lupus erythematosus arthritis.8

AI and computational imaging are transforming dermatology and oncology by enhancing the ability to forecast

disease progression and response to treatments. 9, 10 The progression of such technology in image analysis has

significantly refined the process, enabling intricate pattern detection and forward-looking analysis. 11 This

evolution has revolutionized traditional image processing approaches, providing healthcare professionals with

state-of-the-art, precise, and time-efficient analytical tools. 12 Despite these advancements, such technologies have

not yet been applied to detect inflammatory arthritis through infrared imaging in children.

This study integrates thermal imaging with a computer vision-based analysis protocol, setting the stage for a

transformative diagnostic modality in pediatric rheumatology. By exploiting the recently developed smartphone-

2
based thermal camera and a custom-developed Python script for precise image analysis, we aimed to identify the

distinctive thermal patterns of joint inflammation. The goal is not only to ascertain the presence of inflammation

but also to distinguish the subtle physiological differences between inflammatory joints in afflicted children and

those in their healthy counterparts. This differentiation is pivotal for early therapeutic intervention, precise

monitoring, and ultimately, improving the long-term outcomes for children battling rheumatic diseases.

Materials and Methods

i. Study Population:

In this prospective observational study, we enrolled a cohort of 18 pediatric patients aged 1 to 18 years who were

experiencing an acute knee arthritis flare. Each patient's diagnosis conformed to the stringent criteria established

by the International League of Associations for Rheumatology (ILAR) and Paediatric Rheumatology

International Trials Organization (PRINTO). A control group was established, consisting of 19 age-matched

children from 1 to 18 years, who showed no clinical or laboratory signs of inflammation. Patient’s data, including

age, gender, duration of disease, the specific rheumatic diagnosis, previous and current treatments (NSAIDs,

corticosteroids, or DMARDs), and relevant laboratory markers such as erythrocyte sedimentation rate (ESR), C-

reactive protein (CRP), fibrinogen, and white blood cells (WBC), were collected. Exclusion criteria for all

participants included a history of peripheral arterial disease, autonomic neuropathy, the recent application of

cosmetic products or anti-inflammatory creams, acupuncture, or extremity treatments with significant

temperature variations.

ii. Thermography Protocol

Participants underwent a standard protocol in a dedicated examination room set at 21°C±2·0°C with 50±5%

relative humidity and controlled airflow. Room conditions were monitored using a ThermoPro TP-50 digital

thermometer. After a 20-minute acclimatization period, thermal imaging was conducted using a FLIR ONE Edge

Pro wireless thermal camera, following the best practice guidelines of the Glamorgan Protocol. 6 The FLIR ONE

Edge Pro, with an object temperature range of -10°C to 50°C, thermal resolution of 160 x 120 (19,200 pixels),

thermal sensitivity of <70 mK, and ±5% accuracy, connected wirelessly to smartphones. The camera utilized

VividIR™ and FLIR MSX® technology, ensuring clear thermal images from any angle. The thermal camera was

wirelessly connected to the OPPO A74 (model no CPH2219) smartphone. The thermal camera's emission level

was consistently set to 0·98 for all participants. Thermal images of each participant's affected and unaffected knee

3
were captured from a fixed distance of 1 meter, with the camera mounted on a tripod to ensure stability and

uniformity. The standardized setup allowed for comparable results, as five images were taken at a 90-degree

angle to each subject's knees. All thermal images were automatically saved to the mobile storage folder and FLIR

Ignite software. In the pattern of infrared thermal imagery, the highest temperatures are signified by areas of

white and red, which progressively decrease through shades of yellow and green, culminating in blue as the

representation of the lowest temperatures.

iii. Image Analysis Protocol

The foundation of our analytical framework is a custom-developed Python script that seamlessly integrates

sophisticated computer vision functionalities enabled by OpenCV and the substantial data manipulation

capabilities of pandas (for code specifics, please refer to the supplementary files).13 The OpenCV library was

instrumental in providing interactive image analysis, while Numpy underpinned these processes with its

advanced mathematical computation capabilities. Pandas was indispensable for efficient data management, and

Matplotlib was utilized for the graphical representation of our findings.

The protocol instructs the user to pinpoint two centers for each region of interest (ROI) and a third point on the

ROI's periphery to establish its radius. The script automatically calculates the distance from the center to this

peripheral point to determine the circular ROI's radius (see Fig. 2). It then computes the mean pixel values within

these ROIs by masking the selected areas and averaging the intensities of the color bands—blue, green, and red—

which correspond to temperature readings in infrared images. The script logs the coordinates, radii, and mean

pixel values for each color channel of the ROIs and stores this information in a structured DataFrame (see

supplementary files). To promote accuracy and reproducibility, the script includes feedback mechanisms: It

refreshes the image after each selection to show the current ROIs with their specified radii. This visual

confirmation enables users to verify or modify their selections, thus minimizing the potential for selection errors.

Additionally, FLIR Ignite software enabled us to record the maximum (T max), minimum (T min), and average

temperatures (T ave) for each ROI.

iv. Statistical Analysis and Data Integrity

The study employed both parametric (t-test for normally distributed variables) and nonparametric (Mann-

Whitney U test for non-normally distributed variables) statistical methods to compare the temperature

distributions and means of pixel values between the patient and control groups, respectively. Cohen's d and Effect

4
Size r were calculated to determine the magnitude of differences observed. Diagnostic accuracy was critically

assessed through Receiver Operating Characteristic (ROC) analysis, delineating the Area Under the Curve

(AUC), and computing sensitivity, specificity, as well as predictive values at optimal threshold temperatures. The

Type I error probability was determined as α= .05 in all hypothesis tests, and statistical evaluations were

performed using the SPSS v25·0 software package (IBM Co.).

Result:

In Table 1, the comparison between arthritis patients (n=18) and healthy controls (n=19) reveals no statistically

significant differences in age, with a mean age of 9·46 ± 4·62 years for patients and 11·57 ± 4·86 years for

controls (p=0·183). Gender distribution showed a higher proportion of males in the healthy group (68·4%)

compared to the arthritis group (38·9%), though this difference approached but did not reach statistical

significance (p=0·070). In terms of anthropometric measures, there was no statistical significance in the weight

and height between the groups (weight: 38·04 [13·0-105·0] vs. 50·0 (13·0-85·0) Kg, p = 0·162 and (height:

145·36 ± 31·43 vs. 126·33 ± 30·74 cm, p=0·071). Vital signs including heart rate and blood pressure showed no

significant differences (p>0·05).

In terms of diagnosis, JIA present in 88·9% and cSLE in 11·1% of the patient group, while all healthy controls

were asymptomatic. Arthritis was observed in the knees of all patients: 61·1% had it in the right knee, 55·6% in

the left knee, and 16·6% were affected in both knees. No knee arthritis was reported in healthy controls, as

anticipated.

In terms of infrared temperature distributions, the mean T max of the right knee in arthritis patients was 33·37°C,

which was significantly higher than that of the healthy children (31·73°C, p = 0·001). Similar trends were

observed for the T min and T ave of the right knee (p = 0·014 and p = 0·003, respectively) and for all temperature

measures of the left knee, with p-values of 0·001 for T max, 0·036 for T min, and 0·001 for T ave. The

temperature differences between arthritis patients and healthy children for both right and left knees were

statistically significant (p < 0·05), with effect sizes (Cohen's d) indicating a medium to large effect across the

various measurements (see Table 2).

5
Table 1: Demographic, physiological, and clinical characteristics of arthritis patients compared to healthy
control.

Arthritis patients
Features Healthy children (n=19) p value
(n=18)
Age, mean ± SD (years) 9·46 ± 4·62 11·57 ± 4·86 0·183
Gender – n (%)
0·070
Male 7 (38·9) 13 (68·4)
Girl 11 (61·1) 6 (31·6)
Weight, med (min-max) (kg) 38·04 (13·0-105·0) 50·0 (13·0-85·0) 0·162
Height, mean ± SD (cm) 126·33 ± 30·74 145·36 ± 31·43 0·071
Heart rate, mean ± SD (min) 107·28 ± 16·02 99·86 ± 14·68 0·150
Blood pressure
Systolic, mean ± SD (mmHg) 97 ± 8 98 ± 8 0·635
Diastolic, mean ± SD (mmHg) 65 ± 6 65 ± 5 0·766
Diagnosis – n (%)
JIA 16 (88·9) -
cSLE 2 (11·1) -
Asymptomatic 0 19 (100)
Knee arthritis – (%) 18 (100) -
Right 11 (61·1) -
Left 10 (55·6) -
Both 3 (16·6) -

6
Table 2: Combined table of thermal imaging and color analysis in arthritis patients vs· healthy children

Arthritis patients (n=18) Healthy children (n=19) P

Joint (°C) Cohen's d
Mean ± SD Mean ± SD value

Right knee
T max 33·37 ± 1·57 31·73 ± 1·23 0·001 1·16
T min 31·30 ± 1·73 29·95 ± 1·42 0·014 0·85
T ave 32·38 ± 1·67 30·82 ± 1·25 0·003 1·05
Left knee
T max 33·46 ± 1·42 31·73 ± 1·17 0·001 1·32
T min 31·38 ± 1·73 30·30 ± 1·27 0·036 0·71
T ave 32·54 ± 1·52 30·94 ± 1·17 0·001 1·79
Mann-Whitney U Test
Mann-
Arthritis patients Healthy children Asymp· Sig· (2- Effect
Parameter Whitney
med (min-max) med (min-max) tailed) Size r
U
Right Knee
Blue 61·9 (12·8-140·5) 38·1 (13·2-158·9) 157·00 0·671 0·07
Green 100·6 (30·6-196·9) 175·9 (9·3-200·3) 90·00 0·014 0·40
Red 227·5 (170·8-253·1) 175·4 (28·3-242·4) 80·00 0·006 0·50
Left Knee
Blue 62·9 (17·9-143·5) 31·4 (13·4-155·7) 131·00 0·224 0·20
Green 91·3 (23·1-202·2) 171·4 (27·7-201·5) 100·00 0·031 0·35
Red 228·2 (124·7-242·5) 190·7 (21·2-251·3) 79·00 0·005 0·50

*Cohen’s d values 0·2, 0·5, and 0·8 represent small, medium, and large effect size, respectively. **Effect size r
values 0·1, 0·3 and 0·5 represent small, medium, and large effect size, respectively. The blue, green, and red
color bands represent the mean pixel values from infrared thermography.

In our study's detailed analysis of infrared thermography, significant differences were observed in the color

spectrum between children with knee arthritis and their healthy counterparts. Specifically, the median pixel

values within the red spectral band – a critical indicator of heat due to inflammation – showed marked disparities.

For the right knee, the median pixel value in the red band for patients was 227·5, with a range from 170·8 to

253·1, significantly higher than the median of 175·4 for healthy children, whose values ranged from 28·3 to

242·4. This difference was not only statistically significant, as evidenced by a Mann-Whitney U test resulting in a

7
p-value of 0·006, but it also showed a notable effect size of r=0·50, suggesting a strong correlation. Similar

distinctions were found in the left knee, with patient values at a median of 228·2 (ranging from 124·7 to 242·5)

versus 190·7 (ranging from 21·2 to 251·3) for healthy controls, also yielding a significant p-value of 0·005 and

an effect size of r=0·50. These findings, outlined in Table 2, highlight the robust diagnostic potential of the red

spectral band in thermographic images for detecting inflammation in pediatric arthritis.

In the evaluation of diagnostic accuracy for knee assessments, two indicators were considered: T max and T ave.

The Receiver Operating Characteristic (ROC) analysis yielded Area Under the Curve (AUC) values of 0·822 for

knee T max and 0·797 for knee T ave, indicating a strong diagnostic ability for both parameters. The optimal

threshold value determined for knee T max was 32·25°C, with a high sensitivity rate of 86·1% suggesting that the

test is adept at identifying individuals with the condition. However, the specificity was 71·1%, indicating a

moderate rate of correctly identifying those without the condition. The positive predictive value (PPD) was high

at 86·1%, and the negative predictive value (NPD) was 71·1%, implying a reliable predictive capacity for

positive and negative test results respectively. For knee T ave, the threshold value was slightly lower at 31·60°C.

It showed a sensitivity of 75·0%, indicating good detection of positive cases. The specificity was higher than that

of Knee T max at 76·3%, suggesting a slightly better performance in correctly ruling out the condition in non-

affected individuals. The positive predictive value (PPD) and negative predictive value (NPD) were 77·8% and

78·9% respectively, reflecting a balanced accuracy in predicting true positives and true negatives. These findings

are visually supported by the ROC Curve in Figure 2 and quantitatively detailed in Table 3, underscoring the

robustness of thermal measurements in diagnosing pediatric knee arthritis.

Table 3: Diagnostic accuracy parameters for knee arthritis using temperature assessment.

Cut-off,
Temp °C AUC (95% CI) Sens. (%) Spec. (%) PPD (%) NPD (%)
°C
Knee
T max 0·822 (0·720-0·924) 32·25 86·1 71·1 86·1 71·1
T ave 0·797 (0·690-0·904) 31·60 75·0 76·3 75·0 76·3

Discussion:

Our study provides valuable insights into infrared thermography's potential, coupled with sophisticated computer

vision algorithms, as a non-invasive diagnostic method for pediatric rheumatologic conditions. We identified

8
statistically significant differences in infrared temperature measurements and color analysis between children

with arthritis and their healthy peers, underscoring infrared technology's utility in clinical practice.

Previous research has established thermal imaging as effective in identifying abnormal thermal patterns

indicative of a range of medical conditions, including JIA, RA, diabetic foot, breast cancer, and hidradenitis

suppurativa (HS), with noted thermographic patterns in larger joints in adult RA patients. 14-17 Our study utilized

FLIR ignite to calculate T max, T min, and T ave, revealing that the knees of patients with arthritis exhibited

higher temperature readings in comparison to the knees of healthy subjects. This corroborates existing research

indicating that infrared thermography can discern skin temperature fluctuations stemming from underlying

inflammatory processes.

Traditional investigations into thermal imaging have relied on the proprietary software that comes with thermal

cameras.14-17 Unlike these previous studies, our research leverages automated computer vision to detect and

analyze ROI within thermal images (Figure 2). We found that the blue, green, and red (BGR) pixel values were

significantly higher in arthritis patients compared to healthy children, with the analysis of the color spectrum

further delineating the two groups. Notably, within the red spectral band—which is associated with higher

temperatures indicative of inflammation—the differences were significant, with a substantial effect size (r=0·50)

for the median pixel values, suggesting that thermographic color metrics can be reliable biomarkers for

inflammation. In infrared thermography, the colors in an image serve as visual indicators of temperature

variances, with a commonly used palette ranging from white for the hottest regions to blue for the coldest,

passing through red and yellow for intermediate temperatures. This color coding facilitates quick visual analysis

but should be interpreted with reference to the specific thermal camera's settings and the provided color-

temperature scale to ensure accuracy.6, 18

Bauer et al. recently applied AI-based image processing to thermographic images of 212 young female volunteers

to categorize stages of cellulite with more than 80% accuracy. 10 Additionally, a separate study focused on the

diagnosis of RA in women used manual or semi-automated techniques for thermal image analysis. 19 Our

approach is unique in employing real-time, automated image analysis tailored for infrared image assessment, with

reported effect sizes (r) indicating a medium to large clinical impact. This study introduces a new application of

infrared thermography for the diagnosis of acute knee arthritis in pediatric patients, setting it apart from adult RA

detection methods such as RANet and QNN by targeting localized inflammation in a younger demographic. 20

9
In 2020, Stafford and colleagues conducted a comprehensive review of literature on AI applications in

autoimmune diseases, noting that multiple sclerosis (MS) and adult RA had the highest number of dedicated

studies, with 41 and 32 manuscripts respectively. 21 Adult RA research also demonstrated the broadest spectrum of

input data sources among all autoimmune diseases studied, highlighting the considerable potential for AI

application in this area. However, such systematic and expansive research into AI applications is not as common

in pediatric populations. The unique disease expressions and treatment responses in children warrant more

focused studies to harness the potential of AI in pediatric autoimmune conditions, which could significantly

enhance diagnostics and treatment personalization for younger patients.

Our method's high sensitivity and specificity are evidenced by AUC values of 0·822 and 0·797 for knee T max

and T ave, respectively, and are particularly promising for pediatric screening purposes. However, moderate

specificity (71·1%) for the same threshold value indicates the potential for false positives, which could lead to

unnecessary further testing. The balanced predictive values (PPV and NPV) for T ave suggest that this parameter

may provide a reliable diagnostic criterion when assessing for the absence or presence of inflammation.

Unique to our study is the customized Python script with OpenCV integration for automated ROI detection and

analysis in thermal imagery, minimizing human error and increasing the robustness and reproducibility of our

results. In conclusion, our findings validate the efficacy of infrared thermography, enhanced by computer vision,

for the sensitive detection of joint inflammation in pediatric rheumatologic diseases. While the approach requires

further refinement and validation, it shows promise in streamlining diagnosis, facilitating timely and

individualized treatment. Future research should aim at multicenter and longitudinal studies to bolster these

findings, promoting infrared thermography from a research tool to a routine clinical application, thereby

advancing care in pediatric rheumatology.

10
Refrences:

1. Hootman JM, Helmick CG, Barbour KE, et al. Updated Projected Prevalence of Self-Reported Doctor-

Diagnosed Arthritis and Arthritis-Attributable Activity Limitation Among US Adults 2015-2040. Arthritis

Rheumatol 2016;68(7):1582-7.

2. Bhattad S, Mohite RS, Singh N. Growth and Development in Children with Rheumatic Diseases:

Maintaining a Balance between Drugs and Disease Activity. Indian Journal of Rheumatology 2022;07.

3. Prakken B, Albani S, Martini A. Juvenile idiopathic arthritis. Lancet 2011;377(9783):2138-49.

4. Garner AJ, Saatchi R, Ward O, et al. Juvenile Idiopathic Arthritis: A Review of Novel Diagnostic and

Monitoring Technologies. Healthcare (Basel) 2021;9(12):1683.

5. Giancane G, Consolaro A, Lanni S, et al. Juvenile Idiopathic Arthritis: Diagnosis and Treatment. Rheumatol

Ther 2016;3(2):187-207.

6. Ammer K. The Glamorgan Protocol for recording and evaluation of thermal images of the human body.

Thermol Int 2008;18(4):125-44.

7. Lahiri BB, Bagavathiappan S, Jayakumar T, et al. Medical applications of infrared thermography: a review.

Infrared Phys Technol 2012;55(4): 221-235.

8. Lasanen R, Piippo-Savolainen E, Remes-Pakarinen T, et al. Thermal imaging in screening of joint

inflammation and rheumatoid arthritis in children. Physiol Meas 2015;36(2):273-82.

9. Wilczyński S, Koprowski R, Deda A, et al. Thermographic mapping of the skin surface in biometric

evaluation of cellulite treatment effectiveness. Skin Res Technol 2017;23(1):61-69.

10. Bauer J, Hoq MN, Mulcahy J, et al. Implementation of artificial intelligence and non-contact infrared

thermography for prediction and personalized automatic identification of different stages of cellulite. EPMA

J 2020;11(1):17-29.

11. Aeffner F, Wilson K, Martin NT, et al. The Gold Standard Paradox in Digital Image Analysis: Manual

Versus Automated Scoring as Ground Truth. Arch Pathol Lab Med 2017;141(9):1267-1275.

12. Guan S, Mehta B, Slater D, et al. Rheumatoid Arthritis Synovial Inflammation Quantification Using

Computer Vision. ACR Open Rheumatol 2022;4(4):322-331.

13. Mar CC, Zin TT, Tin P, et al. Cow detection and tracking system utilizing multi-feature tracking algorithm.

Sci Rep 2023;13(1):17423.

11
14. Giancane G, Consolaro A, Lanni S, et al. Juvenile Idiopathic Arthritis: Diagnosis and Treatment. Rheumatol

Ther 2016;3(2):187-207.

15. Ahn SM, Chun JH, Hong S, et al. The Value of Thermal Imaging for Knee Arthritis: A Single-Center

Observational Study. Yonsei Med J 2022;63(2):141-147.

16. Derruau S, Renard Y, Pron H, et al. Combining Magnetic Resonance Imaging (MRI) and Medical Infrared

Thermography (MIT) in the pre- and peri-operating management of severe Hidradenitis Suppurativa (HS).

Photodiagnosis Photodyn Ther 2018;23:9-11.

17. Gatt A, Mercieca C, Borg A, et al. A comparison of thermographic characteristics of the hands and wrists of

rheumatoid arthritis patients and healthy controls. Sci Rep 2019;9(1):17204.

18. Negishi T, Abe S, Matsui T, et al. Contactless Vital Signs Measurement System Using RGB-Thermal Image

Sensors and Its Clinical Screening Test on Patients with Seasonal Influenza. Sensors (Basel)

2020;20(8):2171.

19. Alarcón-Paredes A, Guzmán-Guzmán IP, Hernández-Rosales DE, et al. Computer-aided diagnosis based on

hand thermal, RGB images, and grip force using artificial intelligence as screening tool for rheumatoid

arthritis in women. Med Biol Eng Comput 2021;59(2):287-300.

20. Ahalya RK, Almutairi FM, Snekhalatha U, et al. RANet: a custom CNN model and quanvolutional neural

network for the automated detection of rheumatoid arthritis in hand thermal images. Sci Rep

2023;13(1):15638.

21. Stafford IS, Kellermann M, Mossotto E, et al. A systematic review of the applications of artificial

intelligence and machine learning in autoimmune diseases. npj Digit Med 2020;3:30.

Figure 1: This figure illustrates the efficacy of smartphone IR technology in detecting inflammation in joints

affected by JIA. The depicted thermal images, processed through a dedicated application, highlight areas of

increased temperature that correspond to inflammatory activity within the joint.

Figure 2: Panel (a) displays a flowchart describing the sequence of steps in a Python image processing script. It

begins with 'Read Image', where images are imported using PIL or OpenCV libraries. Following this, the 'Process

Image' step uses NumPy for tasks such as slicing and masking. Next, the script 'Computes Mean Pixel' values by

employing NumPy's mean function. The final step is 'Save/Plot Data', which utilizes Matplotlib to either display

12
or save the processed data. The flowchart concludes with the 'End' of the script. Panel (b) presents a thermal

image of legs, with circles marking the ROIs on the knees. The Python script facilitates the entire process from

importing the thermal image, selecting ROIs, to analyzing the color data automatically.

Figure 3: This ROC curve illustrates the diagnostic performance of knee T max and T ave temperature

assessments in distinguishing between affected and unaffected individuals. The AUC for T max (blue line) and T

ave (green line) confirms the strong diagnostic capabilities of both parameters. The reference line represents a

test with no diagnostic ability (AUC = 0·5).

13