Non-small cell lung cancer (NSCLC) remains a major clinical challenge, with Programmed death-ligand 1 (PD-L1) expression serving as a crucial biomarker to guide immunotherapy. However, its current assessment through invasive biopsies may not capture tumor heterogeneity. This study explores the feasibility of a CT-based radiomics approach, combined with machine learning (ML), as a potential non-invasive virtual biopsy to predict high PD-L1 expression (\ (\) 50%) in NSCLC patients. Contrast-enhanced CT scans from 55 patients with histologically confirmed NSCLC were retrospectively analyzed. Radiomic features were extracted from tumor volumes, and multiple ML classifiers were trained and evaluated through repeated stratified k-fold cross-validation. Among the models evaluated, the Support Vector Machine (SVM) classifier demonstrated the best performance, achieving a median accuracy of 0. 77 (quartiles: 0. 66–0. 82) and an area under the curve (AUC) of 0. 83 (0. 63–0. 92). Feature importance analysis using SHAP (Shapley Additive Explanations) revealed that texture features were the most informative in predicting PD-L1 expression levels. Notably, the integration of clinical data did not improve model performance, highlighting the dominant predictive value of radiomic features alone. Our findings support the feasibility of CT-based radiomics as a potential tool for virtual biopsy to identify NSCLC patients with high PD-L1 expression (\ (\) 50%), potentially serving as a complementary or alternative tool to tissue biopsy, especially in cases where biopsy is contraindicated or insufficient.
Destito et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: