Integrating Tumor and Nodal Image Features at Baseline and Mid-treatment PET Scans to Predict Early Progression in Stage III Non-Small Cell Lung Cancer (NSCLC) Patients Treated with Concurrent Chemoradiotherapy

Abstract

Prognostic biomarkers that can reliably predict early disease progression in stage III non-small cell lung cancer are needed for identifying those patients at high risk of treatment failure, who may benefit from more intensive treatment strategy. This work aims to identify an imaging signature in predicting progression in local advanced NSCLC. This retrospective study included 82 patients treated with definitive chemoradiotherapy for whom both baseline and mid-treatment PET/CT scans were available. All primary tumors and involved nodes were delineated. Patients were randomly divided into two groups: training group (n=41) and testing group (n=41). For training group, forty-five quantitative image features were extracted to characterize the tumor and involved lymph nodes from PET scans at both time points as well as changes between two scans. These features included volume, intensity and texture features for both tumor and nodes. For nodes, there were three additional features, including number of nodes, node-tumor spread which means the maximum distance between tumor and involved nodes, nodal spread which means the maximum distance between involved nodes. Random survival forest (RSF) model with repeated 10-fold cross validation was implemented for predicting progression-free survival (PFS). At last, the RSF model was validated with the testing group. The contribution of individual image features to the prediction was obtained from the RSF model. The most important feature for predicting PFS was the pre-treatment tumor volume, following by the changes of maximum distance between tumor and involved nodes measured at two time points and then the node to tumor spread measured at mid-treatment. Nodal spread was both important at baseline and mid-treatment. Mid-treatment tumor volume was also useful, which was ranked at 5th. In multivariable analysis, the RSF model was an independent prognostic factor for patients’ progression-free survival (HR= 1.17, 95% CI: 1.05-1.31, P= 0.005). In training group, the median risk score based on the RSF model can significantly stratify patients into low- vs. high-risk groups for suffering disease progression (P<0.001, log-rank test). Using the same cutoff value, the RSF model was validated by the testing group (P= 0.016,log-rank test). Integration of tumor and nodal imaging characteristics at pre and mid-treatment PET scans may allow better prediction of progression-free survival for local advanced NSCLC patients, and help to optimize radiation intensity or modify the therapeutic regimen.