PET/CT Radiomics-based assessment of tumor heterogeneity for predicting immunotherapy outcomes in advanced lung...

Accurately predicting progression-free survival (PFS) in patients with advanced lung squamous cell carcinoma (LUSC) receiving immunotherapy remains a clinical challenge. Radiomics has emerged as a promising non-invasive approach; however, its application in this specific patient population remains relatively limited.

A total of 129 patients were retrospectively enrolled. Radiomics features were extracted from baseline positron emission tomography/computed tomography (PET/CT) images. Feature reproducibility was evaluated using intraclass correlation coefficients (ICC), and only features with ICC > 0.75, indicating good reproducibility, were retained. A three-step feature selection strategy, including univariate Cox regression, least absolute shrinkage and selection operator (LASSO), and multivariate Cox modeling, was applied to construct the radiomics signature. Model performance was evaluated using time-dependent receiver operating characteristic (ROC) analysis, Kaplan-Meier survival analysis, and bootstrap resampling. Additionally, a 3D convolutional neural network (CNN) model was explored for comparative analysis.

The final model achieved a 2-year AUC of 0.673. The relatively stable AUC values across time points (1-year: 0.670; 2-year: 0.673; 3-year: 0.651) suggest that the model primarily captures baseline risk stratification rather than strong time-dependent variation. Patients in the high-risk group exhibited a significantly increased risk of progression compared with those in the low-risk group (HR = 4.36, 95% CI: 2.70-7.04, P < 0.001). The radiomics signature remained an independent predictor of PFS in multivariate analysis. In comparison, the 3D CNN model demonstrated inferior predictive performance compared with the radiomics-based model.

The PET/CT-based radiomics model enables non-invasive risk stratification of patients with advanced LUSC receiving immunotherapy and may provide complementary information for clinical decision-making.