Novel MRI-based quantitative signal-intensity ratio scores for differentiation of pyogenic and tuberculous...

This study aimed to develop quantitative MRI-based signal-intensity ratio scores, namely the Epidural Abscess Signal (EAS) score and Vertebral Body Signal (VBS) score, to differentiate pyogenic spondylitis (PS) from tuberculous spondylitis (TS) and identify their optimal diagnostic thresholds.

This retrospective study included patients with a definitive diagnosis of PS or TS between December 2021 and December 2024. EAS scores were calculated from T2-weighted imaging signal intensity, and VBS scores were calculated from T1-weighted imaging signal intensity. Receiver operating characteristic (ROC) analysis was used to identify the optimal diagnostic thresholds and evaluate their diagnostic performance for differentiating PS from TS.

A total of 88 patients with confirmed diagnoses of spondylitis (39 PS, 49 TS) were included in the study. Significant differences were observed in EAS scores (0.534 ± 0.132 vs. 0.354 ± 0.092) and VBS scores (1.677 ± 0.454 vs. 2.312 ± 0.534) between PS and TS (both p < 0.001). ROC analysis revealed that the area under the curve (AUC) was 0.888 for EAS and 0.870 for VBS, both showing higher AUCs than individual laboratory biomarkers. The optimal diagnostic thresholds for predicting PS were EAS ≥ 0.400 and VBS ≤ 2.000, with sensitivities of 93.1% and 89.7%, and specificities of 85.3% and 89.8%, respectively. Univariate analysis identified eight variables (p < 0.05) included in multivariable logistic regression model, which revealed that VBS score ≤ 2.000 (adjusted OR = 298.96) and CRP levels ≥ 50.00 mg/L (adjusted OR = 46.37) were independently associated with PS in the entire cohort. Subgroup analysis of patients with epidural abscess demonstrated that EAS ≥ 0.400 (adjusted OR = 62.18) and VBS ≤ 2.000 (adjusted OR = 51.52) remained significantly associated with PS.

This study developed MRI-based quantitative signal-intensity ratio scores (EAS ≥ 0.400 and VBS ≤ 2.000) to differentiate PS from TS, achieving diagnostic accuracies of 88.9% and 89.8%, respectively.