Multiparametric MRI Assessment of Cervical Lymphadenopathy: Combined Diagnostic Performance of Morphological Features...

Background: Accurate differentiation between benign and malignant cervical lymphadenopathy remains clinically important for diagnostic stratification and treatment planning. This study evaluated the diagnostic performance of conventional morphological magnetic resonance imaging (MRI) features and apparent diffusion coefficient (ADC) values in a mixed cohort of cervical lymphadenopathies. Methods: This retrospective lesion-based diagnostic study included 88 cervical lymph nodes from 39 patients who underwent head-and-neck MRI between September 2023 and December 2025. The cohort had malignant entities such as squamous cell carcinoma metastases, thyroid carcinoma, non-Hodgkin lymphoma, adenoid cystic carcinoma, and medullary thyroid carcinoma, as well as benign/reactive, inflammatory, CMV-related, tuberculous, Warthin tumor-associated, and cystic lymphangioma-related lymphadenopathies. MRI examinations were performed for heterogeneous indications, including the initial assessment of palpable cervical lymphadenopathy, oncological staging, post-biopsy follow-up, suspected recurrence, and benign/inflammatory lesion characterization; therefore, not all patients underwent MRI for the same clinical indication. Most examinations were performed during the initial diagnostic work-up, while six cases represented post-biopsy follow-up. Morphological features and ADC values were analyzed using Mann-Whitney U tests, chi-square tests, ROC analysis, DeLong testing, Firth penalized logistic regression, generalized estimating equations (GEE), patient-level bootstrap resampling, and calibration analysis. Statistical analyses were performed using Python (Version 3.12), with exploratory verification in JASP. Statistical significance was set at p < 0.05. Results: The cohort included 39 patients with a mean age of 54 years (range: 18-74 years), with 20 males and 19 females. Of the 88 lymph nodes, 33 were malignant and 55 benign. Malignant nodes demonstrated significantly lower ADC values than benign nodes (0.87 ± 0.23 vs. 1.25 ± 0.22 × 10^-3 mm²/s; U = 207, p < 0.001). ADC alone showed good diagnostic performance, with an AUC of 0.886 (95% CI: 0.803-0.960). The optimal ADC cutoff was 0.900 × 10^-3 mm²/s, yielding 75.8% sensitivity and 89.1% specificity. The final GEE model, including the ADC, nodal shape, and margin characteristics while accounting for intra-patient clustering, achieved an apparent AUC of 0.956. Leave-one-patient-out cross-validation yielded an AUC of 0.929, and the bootstrap optimism-corrected AUC was 0.949. DeLong testing confirmed that the combined model significantly outperformed the ADC alone (AUC improvement = 0.070; p = 0.006), and the inter- and intra-observer reproducibility for ADC was excellent. Conclusions: ADC values, nodal shape, and margin characteristics provide complementary diagnostic information for differentiating benign from malignant cervical lymph nodes. A structured multiparametric MRI approach demonstrated high diagnostic performance, although the findings should be interpreted in the context of the retrospective single-center design and histopathological heterogeneity.