2169 - Preoperative PET/CT Personalized Radiomics Index with Deep Learning to Predict the Benefit of Postoperative Radiotherapy in Patients with pN2 NSCLC: A Pilot Study
Presenter(s)
Z. Ma1, and Z. Hui2; 1Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 2Department of VIP Medical Services, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
Purpose/Objective(s): The value of postoperative radiotherapy (PORT) for patients with pN2 non-small cell lung cancer (NSCLC) is debated. A subgroup of these patients may derive significant benefit from PORT. Radiomics based on PET/CT has revealed insights into tumor heterogeneity. Our goal is to identify the patients most likely to benefit from PORT.
Materials/Methods: Patients with pN2 NSCLC after radical surgery between July 1, 2013, and June 30, 2019, were enrolled in one academic institution as the training set. Patients in another independent academic medical center were enrolled as external validation sets. The personalized radiomic index (PRI) was developed using radiomic texture features extracted from the primary lung nodule in preoperative CT scans. We used a least absolute shrinkage and selection operator-Cox regularization model for data dimension reduction and feature selection. PRI was developed using a deep learning model known as DeepSurv. The area under the curve (AUC) of the receiver operating characteristic was used to assess the PRI's prediction ability. We compared the overall survival (OS) of patients who received PORT against those who did not in the subgroups determined by the PRI.
Results: The training and external validation datasets comprised 165 and 170 participants. In the training cohort, PRI predicted OS (AUC = 0.84, 95% CI: 0.75-0.93, for 3-year OS). The high-PRI group had significantly worse OS than the low-PRI group (3-year OS, 29.95% vs. 90.87%, p<0.01). The median PRI was used to categorize individuals into two risk groups. While patients in low-risk groups did not benefit from PORT (3-year OS 100% vs. 95.41%, p = 0.38), patients in the high-risk group showed a benefit trend (3-year OS 67.86% versus 86.67%, p = 0.18). The external cohort demonstrated consistent findings.
Conclusion: We developed a preoperative PET/CT-based PRI with deep learning capable of predicting OS and the benefits of PORT for patients with pN2 NSCLC.