Prediction of lymph node metastases in NSCLC

Abstract

To distinguish between benign and malignant mediastinal lymph nodes in patients with NSCLC by comparing 2D and semi-automated 3D measurements in FDG-PET-CT.FDG-PET-CT was performed in 46 patients prior to therapy. 299 mediastinal lymph-nodes were evaluated independently by two radiologists, both manually and by semi-automatic segmentation software. Longest-axial-diameter (LAD), shortest-axial-diameter (SAD), maximal-3D-diameter, elongation and volume were obtained. FDG-PET-CT and clinical/FDG-PET-CT follow up examinations and/or histology served as the reference standard. Statistical analysis encompassed intra-class-correlation-coefficients and receiver-operator-characteristics-curves (ROC).The standard of reference revealed involvement in 87 (29%) of 299 lymph nodes. Manually and semi-automatically measured 2D parameters (LAD and SAD) showed a good correlation with mean intraclass coefficients of .80 and .72, respectively. Semi-automated prediction revealed the highest areas-under-the-ROC-curve for volume (.75, 95%CI: .69-81) and SAD (.75, 95%CI: .70-.81). AUC for LAD and maximal-3D diameter were about .68. Substantially lower accuracies were found for elongation (.57, 95%CI: .50-.64).Optimized semi-automated three dimensional parameters by CT cannot approximate reported data on FDG-PET-CT for lymph node assessment in NSCLC. SAD remains the most accurate and at the same time simple to achieve anatomical criterion for definition of NSCLC target lesions.