Development and Validation of a CT-Based Intratumoral-Peritumoral Radiomics Signature for Tumor Mutational Burden Prediction with Prognostic Implications in Non-Small Cell Lung Cancer: A Multicenter Study
Huaxin Li, Ruotong Wu, Qingqing Hui, Lu Zhang, Wei Du, Hua Wang, Zhaoxiang YeIntroduction
To develop and validate an intratumoral–peritumoral CT radiomics signature for predicting tumor mutational burden (TMB) in non–small cell lung cancer (NSCLC) and assess its prognostic value in advanced NSCLC treated with immune checkpoint inhibitors (ICIs). The proposed TMB classifier aims to support noninvasive patient stratification and guide clinical decision-making regarding molecular testing and immunotherapy selection.
Methods
In this multicenter retrospective study, the TMB prediction cohort comprised 380 patients (training set, n=228; validation set, n=97; external testing set, n=55), and 65 patients receiving ICIs formed an independent immunotherapy validation cohort. Intratumoral (T-Rad) and peritumoral (P3-, P5-, P7-Rad) models were constructed after feature selection. Each peritumoral model was fused with T-Rad to generate F3-, F5-, and F7-Rad models. Radiomics features were integrated with independent clinical predictors to establish a combined model. The predicted probability (Rad_signature) from the optimal fused model was evaluated for prognostic value. Gene enrichment analysis was performed in 89 NSCLC patients from The Cancer Imaging Archive (TCIA) database to investigate the biological basis of the Rad_signature.
Results
The F5-Rad model significantly outperformed the T-Rad with AUCs of 0.827 (95%CI: 0.761-0.892), 0.785 (95%CI: 0.687-0.882), and 0.756 (95%CI: 0.581-0.931) in the training, validation, and external testing sets (P = 0.017, 0.041, and 0.024). The combined clinical-radiomics model achieved the highest AUCs of 0.888 (95%CI: 0.839-0.937), 0.825 (95%CI: 0.742-0.907), and 0.765 (95%CI: 0.592-0.938), respectively. In the immunotherapy cohort, Rad_signature (HR = 0.483, 95%CI: 0.236-0.990, P = 0.047) served as an independent prognostic factor for progression-free survival and demonstrated good calibration at 6, 12, and 24 months. Gene Ontology analysis highlighted chromosome segregation and mitotic processes, while gene set enrichment analysis identified cell cycle and proliferation-related pathways, particularly E2F targets and G2M checkpoint.
Conclusions
A CT-based radiomics signature enables noninvasive TMB prediction and provides prognostic relevance in NSCLC patients receiving immunotherapy.