CT radiomics predicts high PD-L1 in NSCLC: SVM AUC 0.83

CT-based radiomics with an SVM classifier predicted high PD-L1 expression (≥50%) in NSCLC with AUC 0.83 (IQR 0.63-0.92) in a small retrospective study of 55 patients. Clinical data did not improve performance; texture features were most informative. Not yet externally validated.

• Retrospective analysis of 55 NSCLC patients with contrast-enhanced CT; radiomic features extracted from tumor volumes and evaluated via repeated stratified k-fold cross-validation.
• SVM model achieved median accuracy 0.77 (IQR 0.66-0.82); adding clinical data did not improve prediction.
• Small sample size, single-center, no external validation; performance may not generalize beyond this dataset.

RadPigeon summaries are original and for information only. They are not clinical advice.