A PET-CT Radiomics Model for Predicting Distant Metastasis in Early-Stage Non–Small Cell Lung Cancer Patients Treated with Stereotactic Body Radiotherapy: A Multicentric Study

Abstract

<h3>Purpose/Objective(s)</h3> SBRT is a guideline-recommended treatment for patients with early-stage NSCLC who are medically unfit or unwilling to undergo surgery, while some patients can still develop distant failure. The goal of this work is developing and validating a radiomics model for predicting distant failure in early-stage NSCLC patients treated with SBRT. <h3>Materials/Methods</h3> receiving SBRT were enrolled. In total, 103 CT radiomic features and 103 PET radiomic features were extracted from the pretreatment PET-CT images of each lung lesion. Patients with inoperable NSCLC treated with SBRT and have dicome format imaging available were eligible. A CT-based radiomics signature and a PET-based radiomics signature were established in one dataset and independently validated in other datasets by logistic regression after feature selection. Multivariable logistic regression analysis was then used to develop a radiomics model incorporated in the radiomics signature and independent clinical predictor. The performance of the model was assessed by its discrimination, calibration, and clinical usefulness with internal (first center), and external (other two centers) validation. <h3>Results</h3> A total of 174 patients (with 174 lung lesions; 126, 21, and 27 from three individual institutions) were enrolled. Median follow-up was 30.7 (1.9-97.8) months. Radiomics signature was built by combining features selected from the feature pool. A model of combining CT and PET radiomic signatures with clinically significant factors like gender by logistic regression had AUC of 0.804 in the training set of first center. This model was then tested from data of first center and other two centers which showed AUC 0.785 and 0.712, respectively. <h3>Conclusion</h3> The presented radiomics model, available as an online calculator, can serve as a user-friendly tool for individualized prediction of the distant failure in early-stage NSCLC patients treated with SBRT.