MRI-based interpretable radiomics nomogram for discrimination between Brucella spondylitis and Pyogenic spondylitis

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning in Medicine

Background:

• Pyogenic spondylitis (PS) and Brucella spondylitis (BS) are common spinal infections with similar clinical and imaging features, complicating diagnosis.
• Delayed diagnosis of PS and BS can lead to vertebral destruction, kyphosis, and neurological deficits, underscoring the need for accurate differentiation.
• Distinguishing between PS and BS is challenging in clinical practice, necessitating advanced diagnostic tools.

Purpose of the Study:

• To develop and evaluate a radiomics nomogram utilizing magnetic resonance imaging (MRI) for accurate differentiation between PS and BS.
• To compare the diagnostic performance of a conventional clinical model, a radiomics score, and a composite model integrating both.

Main Methods:

• Retrospective analysis of MRI and clinical data from 133 patients with pathologically confirmed PS (68) and BS (65).
• Development of a clinical model (M1) using logistic regression and a radiomics score (M2) using LASSO regression on features from sagittal fat-suppressed T2-weighted imaging (FS-T2WI).
• Creation of a composite model (M3) by combining M1 and M2, with performance evaluated using ROC curves, calibration, and decision curve analysis.

Main Results:

• A composite model (M3) achieved the highest diagnostic performance with an Area Under the Curve (AUC) of 0.868 in the testing set, outperforming the clinical model (AUC 0.795) and radiomics score (AUC 0.859).
• The composite model demonstrated excellent calibration and clinical utility as indicated by decision curve analysis.
• SHapley Additive exPlanations (SHAP) were used to enhance the interpretability of the model's predictions.

Conclusions:

• A nomogram integrating MRI-derived radiomics features and clinical data significantly improves the accuracy of differentiating between PS and BS.
• This radiomics nomogram offers a promising tool for precise diagnosis in clinical settings, facilitating personalized treatment strategies for spinal infections.