Magnetic Resonance Imaging-Based Radiomic Signatures for the Diagnosis of Zoster-Associated Pain: A Quantitative Imaging Approach

INTRODUCTION: Accurate diagnosis and treatment of zoster-associated pain (ZAP) remains challenging because identifying the affected dorsal root ganglion (DRG) relies on subjective assessment and lacks objective imaging. We developed and validated a quantitative model using neurological magnetic resonance imaging (MRI) and radiomics, with the contralateral healthy DRG as an internal control. This study aimed to identify imaging biomarkers for precise localization, optimized treatment, and improved patient prognosis. METHODS: Clinical data were retrospectively collected from patients with ZAP who underwent neuro-magnetic resonance imaging (MRI) at the Pain Department of the First Affiliated Hospital of Fujian Medical University (December 2023-December 2024). T2-weighted neuroimages were acquired for region-of-interest segmentation. Experienced radiologists manually segmented neuropathic DRG by using the Insight Segmentation and Registration Toolkit Segmentation and Neuroimaging Applications Platform (ITK-SNAP), and quantitative radiomic features were extracted via PyRadiomics. Feature selection was performed using the Mann-Whitney U test and least absolute shrinkage and selection operator. The selected features were used to train Logistic Regression, Support Vector Machine, Neural Network, XGBoost, and AdaBoost models. Performance metrics included area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. RESULTS: Overall, 408 DRGs from 68 patients (30 male, 38 female; mean age 68.25 ± 9.01 years) were analyzed. From 1316 radiomic features per region, 10 robust and discriminative features were retained for model construction. The neural network achieved the highest diagnostic performance in the training and validation set (AUC 0.818; 95% confidence interval: 0.719-0.916, p = 0.001). Decision curve and calibration analyses indicated moderate clinical utility in distinguishing affected from healthy DRGs. CONCLUSIONS: Magnetic resonance imaging-based radiomics using a neural network algorithm robustly discriminates lesioned dorsal root ganglia in zoster-associated pain. This approach provides an objective imaging tool to improve diagnostic precision and guide targeted therapy.