The value of MRI radiomics in distinguishing different types of spinal infections

Abstract

Background

In clinical practice, the three most prevalent forms of infectious spondylitis are tuberculous spondylitis (TS), brucellosis spondylitis (BS), and pyogenic spondylitis (PS). It is possible to successfully lessen neurological and spinal damage by detecting them early. In the medical field, radiomics has been applied extensively. It is crucial to find out if MRI imaging can be used to diagnose spinal infections early.

Purpose

To explore the diagnostic value of establishing models based on MRI radiomics for different spinal infections.

Methods

This retrospective study collected clinical and magnetic resonance imaging information on a total of 136 patients diagnosed with spondylitis in April 2019 and August 2023, who were classified into specific spinal infections (TS or BS) and non-specific spinal infections (PS) based on treatment. 3D Slicer software was used to outline the region of interest (ROI) and extracted ROI features. All patients were randomly divided into a training set and a test set (7:3), and after standardized, the t-test and LASSO were sequentially performed in the training set to extract the optimal radiomic features. These features were used to calculate the Radscore and construct the features classifier model and evaluated by test set. Univariate and multivariate logistic regression of Radscore and clinical features to identify predictors contributing to the diagnosis were used to plot nomograms, the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA) to assess the nomogram. The same approach described above was used to diagnose both subgroups of BS and TS in SSI.

Results

321 radiological features were extracted from the three different sequences. The remaining 7 optimal radiomics features were used to calculate the Radscore and establish three feature classifier models, with RF having the best performance (AUC=1 and 0.86). And after univariate and multivariate logistic regression, the final nomogram constructed by Radscore and had good discriminatory performance in the training set and the test set (AUC =0.924 and 0.868), and the calibration curve and DCA showed good clinical efficacy. In the subgroup, the AUC of the training and test sets was 0.929and0.863.

Conclusion

The diagnostic model based on MR radiomics can gradually differentiate tuberculous spondylitis, brucellosis spondylitis, and pyogenic spondylitis.