Eur J Radiol. 2025 Apr 10;187:112106. doi: 10.1016/j.ejrad.2025.112106. Online ahead of print.
ABSTRACT
OBJECTIVES: This study aims to investigate the value of a radiomics nomogram based on magnetic resonance imaging (MRI) in distinguishing vertebral compression fractures (VCFs) from vertebral osteomyelitis (VOs).
MATERIALS AND METHODS: We conducted a retrospective analysis of the clinical data from 100 patients with VCFs and VOs, respectively at our hospital. The cases were randomly divided into training (n = 140) and testing sets (n = 60) in a 7:3 ratio. Two experienced radiologists outlined the regions of interest (ROI) on the MRI images using T2-weighted fat suppression (T2WI-FS) images and extracted the radiomic features. The Least Absolute Shrinkage and Selection Operator (Lasso) algorithm was used to select and reduce radiomic features to establish a radiomics model (Model 1), and a Logistic Regression algorithm was used to construct a radiomics score. A multivariable logistic regression analysis was conducted to establish a clinical model (Model 2). A combined model (radiomics nomogram, Model 3) was built based on the radiomics score and independent clinical factors. The diagnostic performance of Models 1, 2, and 3 was validated using the Area Under the Curve (AUC) and Decision Curve Analysis (DCA).
RESULTS: The training and testing sets included 68/72 VCFs and 32/28 patients with VOs, respectively. There were no statistically significant differences in clinical characteristics such as age, sex, body mass index (BMI), CRP levels, ESR, and lesion stage between the training and testing sets (P > 0.05). A total of 873 radiomic features and 6 clinical features were extracted. After screening, 10 optimal features were selected to build Model 1, while 5 clinical features were used to build Model 2. Models 1 and 2 were combined to create Model 3 and a nomogram was plotted. All the three models were constructed using Logistic Regression algorithms. Model 3 achieved a higher AUC than Models 1 and 2 for both the training and testing sets: 0.946 > 0.904 > 0.871 (training) and 0.900 > 0.854 > 0.818 (testing), respectively. Additionally, the DCA indicated that Model 3 had better clinical utility than Models 1 and 2.
CONCLUSION: Our analysis indicated that the radiomics nomogram, combined with radiomic and clinical features, provides significant clinical guidance in distinguishing vertebral compression fractures from spinal vertebral osteomyelitis.
PMID:40228322 | DOI:10.1016/j.ejrad.2025.112106
