Background/Purpose: Rapid and accurate detection of arthritis is crucial for initiating timely treatment and preventing long-term joint damage. Imaging techniques such as Ultra sound and MRI are used for detection of subclinical inflammation, yet can be time-consuming, expensive, and require skilled personal. In contrast, thermal imaging offers a non-invasive, radiation-free alternative that can quickly assess temperature variations associated with inflammation. This study explores the application of a hand-held thermal imaging device in conjunction with machine learning algorithms to detect arthritis.  

Methods: A total of 232 participants were recruited for the study. This included 81 healthy individuals and 126 patients with rheumatoid arthritis, psoriatic arthritis, gout , or pseudogout. Additionally, 25 patients with osteoarthritis (OA) participated. Among these, 18 patients underwent MRI to provide a comparative baseline for the thermal imaging data. Thermal images of the knees, hands, and ankles were captured using a FLIR C5 compact thermal camera device (FLIR Systems, Inc., Wilsonville, OR, USA) from both arthritic patients and healthy controls. A novel algorithm for image processing and machine learning was implemented using MATLAB software (MathWorks Inc., Natick, MA, USA). The algorithm extracted temperature and texture features from the images. Statistical analysis was conducted using unpaired t-tests in GraphPad to analyze the temperature and textural features extracted (Entropy, Kurtosis, Skewness, Local entropy) ,from the images.

Results: The analysis showed that the temperature and textural features of thermal images from patients with arthritis (knee, n=20 and hand, n=25) were significantly different (p< 0.05) compared to healthy controls (knee, n=56 and hand, n=50). The textural parameters in the thermal images of the anterior wrist were statistically different between the healthy controls (n=41) and patients with subclinical arthritis, where synovitis was confirmed by MRI (n=18). The machine learning algorithm developed in MATLAB accurately distinguished between the groups with an accuracy greater than 80%.

Conclusion: The study’s findings suggest that thermal imaging can effectively identify inflammation in arthritic patients. The significant differences in temperature and textural features between healthy controls and patients highlight the potential of thermal imaging in early arthritis detection. Machine learning further enhances diagnostic accuracy, providing a reliable tool for distinguishing between different patient groups. This study demonstrates that a hand-held thermal imaging device, combined with advanced machine learning techniques, may serve as a non-invasive, non-intrusive, radiation-free, cost-effective, and easy-to-use tool for the detection of inflammatory arthritis.

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/accurate-detection-of-arthritis-using-hand-held-thermal-imaging-and-machine-learning/