MRI radiomics reveals pain-related biomarkers for modic changes.

BACKGROUND: Modic changes (MC) are vertebral endplate abnormalities detectable on MRI, strongly linked to low back pain (LBP). Identifying imaging features correlated with pain severity may improve diagnosis and guide treatment. This study aims to develop an automated radiomics-based model for classifying MC and investigating its association with LBP severity. METHODS: A total of 304 retrospective and 101 prospective patients with MRI-confirmed MC (January 2020–May 2024) were analyzed. Radiomic features were extracted from T1, T2, and FS sequences using Pyradiomics, yielding 1,906 features. The Mann-Whitney U-test and LASSO algorithm identified 29 key features. A Naive Bayes model was trained (70%) and validated (30%) using retrospective data. Prospective patient data, including Visual Analogue Scale (VAS) and Oswestry Disability Index (ODI) scores, were analyzed for correlations with pain severity. RESULTS: The model achieved 99% accuracy in classifying Modic types. Seven radiomic features showed significant associations with pain severity. Higher VAS scores correlated positively with short-run high gray-level emphasis and size-zone non-uniformity normalized, and negatively with low gray-level zone emphasis (all P < 0.05). ODI scores showed negative associations with 10th-percentile intensity, elongation, and median intensity, and a positive association with skewness (all P < 0.05). Additionally, focused volumetric analysis (mesh volume, voxel volume, surface area) revealed no linear or nonlinear associations with pain (all P > 0.05), indicating that texture-based inflammatory signatures rather than lesion size may better characterize Modic-related pain. CONCLUSIONS: The radiomics-based Naive Bayes model accurately classifies Modic changes and identifies imaging features linked to LBP severity. These findings support improved diagnostics, early intervention, and personalized rehabilitation for MC-related spinal disorders.