Related Experiment Videos
Integrating Lymph Node Metastasis and Programmed Death-Ligand 1 Prediction in Non-Small Cell Lung Cancer From a
Wen Chen1,2,3,4,5,6, Qiufang Liu1,3,4,5,6, Huiling Peng7
1Department of Nuclear Medicine, Fudan University Shanghai Cancer Center, Fudan University, Rd. Dongan 270, Shanghai, Shanghai, 2000027, China, 86 68388096, 86 68388096.
Background:
Preoperative stratification for non-small cell lung cancer (NSCLC) necessitates separate evaluations of lymph node metastasis (LNM) to guide surgical decisions and of programmed death-ligand 1 (PD-L1) expression to inform immunotherapy.
Objective:
This study aimed to develop and validate an integrated diagnostic solution that could simultaneously predict both LNM status and PD-L1 expression status from a single, standard-of-care 18F-fluoro-2-deoxy-D-glucose positron emission tomography/computed tomography (18F-FDG PET/CT) scan.
Methods:
In this multicenter study, we segmented primary tumors and peritumoral 15-mm expansion regions from preoperative PET/CT scans of 273 patients (for LNM prediction) and 242 patients (for PD-L1 prediction). A total of 7868 radiomic features from intratumoral and peritumoral regions were extracted. Following rigorous feature selection, 2 independent models were developed using machine learning and tested on a temporal validation cohort (n=45). Model performance was benchmarked against clinicopathological models and nuclear medicine physicians.
Results:
The integrated model for LNM prediction (PT-IPT-LR) achieved an area under the curve of 0.845 (95% CI 0.716-0.973) in the temporal validation cohort, with a sensitivity of 0.765 (95% CI 0.518-1.000) and a specificity of 0.786 (95% CI 0.602-0.970). The model for PD-L1 expression (PT-IPT-SVM) achieved an area under the curve of 0.776 (95% CI 0.641-0.911) in the temporal validation cohort, with a sensitivity of 0.800 (95% CI 0.609-0.991) and a specificity of 0.650 (95% CI 0.401-0.899). Decision curve analysis confirmed the clinical utility of both models. Critically, we found no significant correlation between the radiomic signatures of LNM and PD-L1, which validates our 2-model approach.
Conclusions:
We present a radiomics framework that noninvasively integrates prediction of LNM and PD-L1 from a single preoperative PET/CT scan. This tool may enable preoperative stratification, potentially optimizing both surgical and systemic treatment planning for patients with NSCLC in a single step.