Abstract
BACKGROUND: Pyogenic spondylitis (PS) and tuberculous spondylitis (TS) are two common types of spinal infections. Given the significant differences in their antimicrobial treatment regimens, early and accurate differentiation is critical for developing appropriate clinical management strategies. The study aimed to develop and validate MRI-based epidural abscess signal (EAS) and vertebral body signal (VBS) scores for differentiating PS from TS, guiding empirical antibiotic therapy before microbiological confirmation. METHODS: This study enrolled patients with definitive etiological or pathological diagnosis of PS or TS from two tertiary centers into two cohorts: a retrospective derivation cohort (n = 88; December 2021-December 2024) and a prospective validation cohort (n = 34; January-June 2025). The EAS score was calculated using T2-weighted imaging signal intensity, while the VBS score was derived from T1-weighted imaging signal intensity. Diagnostic performance of scores was evaluated using receiver operating characteristic (ROC) curve analysis to determine optimal cutoff values. Multivariable logistic regression analysis was performed to develop the prediction model and externally validated through the prospective cohort. RESULTS: PS patients consistently demonstrated significantly higher EAS scores and lower VBS scores compared to TS patients across both derivation and validation cohorts (all p < 0.001). ROC analysis within the derivation cohort established EAS ≥ 0.400 (AUC = 0.888, 95% CI 0.801-0.975) and VBS ≤ 2.000 (AUC = 0.870, 95% C I0.784-0.956) as optimal diagnostic thresholds for PS. Implementing these thresholds to guide empirical therapy in the validation cohort significantly improved clinical management. This intervention resulted in a marked reduction in inappropriate empirical therapy (5.9% vs. 23.4%), shorter mean hospitalization duration (23.4 vs. 28.7 days), and reduced total antibiotic therapy duration (20.2 vs. 25.2 days) compared to the derivation cohort managed by standard guidelines. Multivariate analysis confirmed EAS ≥ 0.400 (aOR = 50.04, 95% CI 3.15-794.06) and CRP ≥ 50 mg/L (aOR = 37.47, 95% CI 1.01-1384.31) as independent predictors of PS in the EAS subgroup. Similarly, VBS ≤ 2.000 (aOR = 84.10, 95% CI 9.81-721.18) and CRP ≥ 50 mg/L (aOR = 10.62, 95% CI 1.11-101.46) were significant independent predictors in the VBS subgroup. CONCLUSION: The EAS and VBS MRI-based scores enable rapid differentiation of pyogenic from tuberculous spondylitis, significantly reducing inappropriate antibiotic use and optimizing therapeutic decision-making.