Can Computer-Aided CT Analysis Redefine Mediastinal Lymph Node Diagnosis?

Abstract

Background: Accurate mediastinal lymph node evaluation is essential in the diagnosis and staging of malignant tumors. Traditional size-based criteria pose challenges, particularly for smaller nodes, necessitating alternative approaches. Objective: To investigate the diagnostic value of volume, calculated diameter, and HU values obtained using computer-aided software in distinguishing between benign and malignant mediastinal lymph nodes on contrast-enhanced and non-contrast thoracic CT. Methods: A retrospective analysis of 103 patients with 172 lymph nodes. Quantitative metrics were derived using the "Vitrea Lung Nodule Analysis" software. Diagnostic performance was assessed using ROC analysis, and statistical associations were examined through univariate and multivariate logistic regression models. Results: Age and male gender were significant predictors of malignancy (p<0.001). While HU values and calculated parameters showed limited diagnostic performance, multivariate analysis revealed significant effects of lymph node diameter and multiple nodes on malignancy classification (p<0.05). The model achieved an AUC of 0.804, with sensitivity and specificity of 0.815 and 0.684, respectively. Conclusion: Despite incorporating entire lymph nodes in quantitative assessments, HU values and volumes demonstrated limited clinical utility. However, advanced imaging and AI-driven three-dimensional analyses may enhance diagnostic accuracy.