BackgroundMediastinal lymph node (LN) staging is routinely performed using PET/CT and EBUS-TBNA. Promising predictive algorithms for lymph nodes have been reported for each technique, both individually and in combination. This study aims to develop a predictive algorithm that combines EBUS, PET/CT and clinical data to provide a probability of malignancy.MethodsA retrospective study was conducted on consecutive patients with non-small cell lung carcinoma staged using PET/CT and EBUS-TBNA. Lymph nodes were identified by level (N1, N2, and N3) and anatomical region (AR) (subcarinal, paratracheal, and hilar). A Standardized Uptake Value (SUV) was determined for each sampled LN. The ultrasound features collected included diameter in the short axis (DSA), morphology, border, echogenicity and the presence of the vascular hilum. A robust logistic regression model was used to construct an algorithm to estimate the probability of malignancy of the lymph node.ResultsA total of 116 patients with a mean age of 66, 93% of whom were men, were included. 358 lymph nodes were evaluated, 51% of which exhibited adenocarcinoma and 35% were squamous, while 14% were classified as non-small-cell lung carcinoma. The model estimated the probability of malignancy for each lymph node using age, DSA, SUVmax, and AR. The Area Under the ROC curve, was 0.89. A user-friendly application was also developed (https://ubidi.shinyapps.io/lymma/.)ConclusionsThe integration of patient clinical characteristics, EBUS features, and PET/CT findings may generate a pre-sampling malignancy probability map for each lymph node. The model requires prospective and external validation.
Read full abstract