Texture-enhanced thermography for joint inflammation detection using consumer-grade thermal cameras

Abstract

Background Early identification of active synovitis is essential for treat-to-target management in inflammatory arthritis. While MRI and ultrasound are highly sensitive, their clinical use is constrained by high costs, operator dependence, and lack of portability. Infrared thermography (IRT) is rapid and contactless, but conventional temperature-based approaches are sensitive to ambient conditions and inter-individual baseline variability, and contralateral comparison is unreliable in bilaterally involved disease. This study evaluates whether consumer-grade thermal cameras, augmented with texture-based features and complementary differencing strategies, can detect and differentiate joint inflammation. Methods Thermal images of knees, ankles, wrists, and metacarpophalangeal (MCP) joints were acquired from 239 participants (167 with inflammatory arthritis, osteoarthritis (OA), or fibromyalgia; 72 healthy controls). Beyond absolute temperature, higher-order texture features (e.g., entropy, skewness) were extracted to quantify spatial heterogeneity in heat distributions. To reduce environmental and inter-individual variability, two normalization strategies were applied: contralateral left–right (L-R) differencing for unilateral disease and a novel anterior–posterior (A-P) differencing for bilateral or subclinical presentations. Group comparisons used the Mann-Whitney U test ( p  < 0.05), and effect sizes were summarized using Cohen’s d . Results Compared with non-differenced features, contralateral differencing markedly strengthened discrimination for unilateral inflammation, with very large effects at the knee ( d  = 2.15) and ankle ( d  = 1.83). Texture features, specifically entropy and skewness, outperformed absolute temperature in detecting latent (subclinical) wrist inflammation via A-P differencing ( d = − 1.11 ). Thermographic signatures also supported differential characterization, distinguishing inflammatory arthritis from OA ( d = − 1.49 ) and fibromyalgia ( d = − 1.17 ). Furthermore, A-P gradients showed anatomical specificity, with inflammation reducing gradients in wrists but increasing them in MCP joints ( d  = 0.64). Conclusion Texture-enhanced thermography, combined with L-R and A-P differencing, shows promise as an accessible approach for identifying active and subclinical joint inflammation. This approach addresses the limitations of bilateral disease in traditional thermography and widens the clinical applicability of IRT screening.