Abstract
BACKGROUND: Diagnosing infectious spondylitis is challenging due to overlapping clinical features and the lack of standardized diagnostic criteria. While magnetic resonance imaging (MRI) and laboratory findings are critical, studies simultaneously analyzing all four major types of infectious spondylitis remain non-existent. The aim of this study is to fill a critical gap in the current literature by providing the first comprehensive comparison of the MRI characteristics and laboratory data for the four major types of infectious spondylitis: pyogenic spondylitis (PS), tuberculous spondylitis (TS), brucellar spondylitis (BS), and fungal spondylitis (FS). Furthermore, the study aims to propose a decision tree model to assist clinical decision-making, particularly in cases where a biopsy may be delayed or unfeasible. This model is designed to facilitate earlier and more targeted interventions, ultimately leading to improved patient outcomes. METHODS: In this retrospective study, we included 117 patients with confirmed infectious spondylitis (37 PS, 36 TS, 23 BS, and 21 FS) and an external test set of 34 confirmed cases. We analyzed MRI sequences including T2-weighted imaging (T2WI), short tau inversion recovery (STIR), and contrast-enhanced T1-weighted images (T1WI). Clinical and radiological features were assessed by two radiologists and two orthopedists. Statistical analysis was conducted using analysis of variance (ANOVA) and Kruskal-Wallis (K-W) tests. Five machine learning (ML) models with 5-fold cross-validation were developed, and an online application was created based on the optimal model. RESULTS: Significant differences (P<0.001) were observed in clinical features [time to diagnosis, fever, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), albumin/globulin ratio (A/G ratio)] and imaging features (vertebral signal on T2WI, extent of destruction, skip lesions, endplate inflammatory reaction line, vertebral intraosseous abscess) among groups. The random forest model was the most accurate, with an area under the curve (AUC) of 0.94 in the training set and 0.92 in the test set. CONCLUSIONS: A predictive model integrating imaging and clinical features effectively differentiates the four major types of infectious spondylitis, enhancing diagnostic accuracy. The online application extends the practical utility of our findings.