A Radiomics Model Predicting Occult Lymph Node Metastasis In Early Stage Non-Small Cell Lung Cancer

Purpose:Occult lymph node metastasis(OLM)in early stage non-small cell lung cancer(NSCLC)refers to the absence of suspicious lymph nodes in the hilum and mediastinum examined by CT and 18FDG PET/CT before surgery,but lymph node metastasis was confirmed by pathology after surgery.The incidence of OLM is about 10%-20%,which is an important cause of regional recurrence in the surgical and stereotactic body radiotherapy patients with clinical N0 lung cancer.This study intends to combine the radiomics parameters of 18F-FDG PET and the chest CT images of breath holding at the end of inspiratory of the same machine and clinical factors to establish and verify a comprehensive nomogram for predicting OLM in early stage NSCLC patients.Methods117 patients with early stage NSCLC(cT1-2N0M0)who underwent anatomical lobectomy and systemic node dissection in Shandong Cancer Hospital Affiliated to Shandong University were retrospectively collected and their clinical and pathological data were recorded.All patients underwent 18F-FDG PET/CT and chest CT scan of breath holding at the end of inspiratory of the same machine before operation.All primary tumors were FDG PET imaging-positive tumors.FDG PET images were imported into CGITA software,with SUV=2.5 as the threshold to outline the region of interest of the primary tumor,and 60 texture parameters were extracted.The chest CT images of the same machine were imported into IBEX software,the primary lesions were identified in the lung window,the area of interest was manually delineated,and 188 radiomics parameters were extracted.CT radiomics parameters were selected by Lasso logistic regression(LLR),and the CT imaging radiomics score(Rad-score)of OLM was constructed.Independent risk factors for OLM were determined by Logistic univariate and multivariate regression analysis.The consistency index(C-index)was used to evaluate the predictive performance of the model in the training and validation group.The decision curve analysis was used to compare the clinical value of the prediction model.ResultsA total of 19 patients(14 in the training group and 5 in the validation group)developed occult lymph node metastases.Primary tumor SUVmax(OR,1.272)and age(OR,0.917)were independent risk factors for OLM in patients with cN0 early stage NSCLC.ROC area under the curve of SUVmax and age for OLM was0.748(95%CI:0.640-0.837)and 0.694(95%CI:0.583-0.791).The area under the curve of OLM predicted by Rad-score was 0.8116(95%CI:0.708-0.888,training group)and 0.860(95%CI:0.751-0.954,validation group).The nomogram based on the primary tumor SUVmax and age,which C-index was 0.804(95%CI:0.69-0.92,training group)and 0.607(95%CI:0.71-0.98,validation group).The combined nomogram based on the primary tumor SUVmax,age and Rad-score has the best prediction performance,and the C-index was 0.864(95%CI:0.77-0.96,training group)and 0.847(95%Cl:0.71-0.98,validation group),respectively.ConclusionA comprehensive nomogram based on the primary tumor radiomics parameters of 18F-FDG PET and chest CT images of the same machine,SUVmax and age can effectively predict OLM in early stage NSCLC patients.