Shandong Cancer Hospital Affiliated to Shandong First Medical University Jinan, Shandong
X. Yin, Y. Cui, J. Wen, X. Meng, and J. Yu; Department of Radiation Oncology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China
Purpose/Objective(s): After stereotactic body radiation therapy (SBRT) for early-stage lung cancer, there are still many unresolved issues regarding the implementation of precise adjuvant therapies. The neglect of occult lymph nodes metastasis (OLNM) is one of the pivotal causes of early NSCLC recurrence after SBRT. We develop and validate a logistic regression model that enable to predict OLNM. Materials/
Methods: 389 patients at two centers diagnosed with lung adenocarcinoma with radiologically node-negative were included. 241 patients were divided into a training set (n=169)and an internal validation set (n=72)at a ratio of 7:3, while another 148 patients from a different center served as an external validation set. Based on the pathological findings following radical surgery, patients were categorized into two cohorts: lymph node-positive (pN+) and negative (pN-) groups. The radiomics features were extracted by open-source software based on Python. LASSO analysis was used to reduce the data dimensionality and select both radiomic and clinical features. Radiomics and clinical features were screened to establish predictive models. Receiver operating characteristics (ROC) curves, area under ROC curves (AUC) and calibration plot were plotted to evaluate the model performance and clinical application. Results: A total of 944 radiomics features were extracted based on volume of interest in CT images. There were five radiomics features (LSFE, LSFT, WHLGI, WHGI, WLGD) and three clinical features (basic diseases, ECOG performance status, maximum diameter of tumor) for OLNM prediction. The AUC of the clinical model were 0.70, 0.74 and 0.72 in the training set, internal and external set, respectively. The radiomics models demonstrated general performance in three sets (AUC: 0.80,0.73,0.74). Moreover, the combined models exhibited superior performance for predicting OLNM (AUC: 0.83, 0.82,0.80). Conclusion: We developed combined radiomics and clinical machine learning model with better performance, which was used to accurate predict OLNM of early lung adenocarcinoma. The prediction results would determine the real stage and guide target designation for this type of lung adenocarcinoma. Prophylactic lymph node irradiation or intensive treatment should also be considered for high-risk groups.
