Manual vs. Automatic Segmentation in Infrared Thermography: Inter-Rater Reliability and a Longitudinal Single-Case Feasibility Illustration of Agreement for Musculoskeletal Rehabilitation Monitoring.

Background: Infrared thermography (IRT) offers a non-invasive means to monitor skin temperature and physiological changes during rehabilitation. However, variability in image segmentation methods limits its broader clinical use. This study assessed the inter-rater reliability of a manual thermal image segmenatation protocol using FLIR Thermal Studio Pro™ and presents a longitudinal single-case feasibility illustration evaluating the agreement between this manual method and automatic segmentation using ThermoHuman®. Methods: Two regions of interest (ROI) were analyzed throughout the study: the frontal knee and vastus medialis oblique (VMO). For the inter-rater reliability analysis, thermal images from 20 professional athletes were used. For the agreement analysis, 50 lower extremity thermal images were captured from a single athlete recovering from a meniscal repair over a four-month rehabilitation period. Intraclass correlation coefficients (ICC (2,1)) were used to assess agreement and reliability. Results: Inter-rater reliability was excellent for all frontal knee measurements (ICC (2,1) = 0.991, 0.995, and 0.964 for maximum, average, and minimum temperature, respectively) and for VMO maximum and average temperature (ICC (2,1) = 0.967 and 0.981, respectively), with VMO minimum temperature as the exception demonstrating good reliability (ICC (2,1) = 0.779). Excellent agreement was found between methods for all frontal knee (ICC (2,1) = 0.967, 0.949, and 0.985) and VMO measurements (ICC (2,1) = 0.971, 0.973, and 0.980) for maximum, average, and minimum temperature, respectively. Conclusion: Manual segmentation using FLIR Thermal Studio Pro™ provides highly reliable results across multiple subjects, supporting its use as a practical and consistent clinical tool. Furthermore, our single-case feasibility illustration suggests it provides comparable results to automatic segmentation using ThermoHuman®, supporting its potential feasibility for accessible rehabilitation monitoring in injured athletes. Further multi-subject research is warranted to confirm the broader applicability of the agreement findings.