Predicting PD-L1 expression status in NSCLC using radiomic analysis of 18 F-FDG-PET/CT images

Lung cancer is the most prevalent malignancy globally, with prognosis and treatment largely influenced by histological and molecular analyses. Molecular features like PD-L1 positivity help identify patients suitable for immunotherapy. However, obtaining histological samples can be challenging and limited. Radiomic analysis of imaging data provides a non-invasive way to characterize tumor heterogeneity and its complex patterns, which may help predict PD-L1 expression. This study investigates the efficacy of radiomics in forecasting PD-L1 status in NSCLC patients using 18 F-FDG-PET/CT images.

In this retrospective study, the primary staging 18 F-FDG-PET/CT scans of 105 patients with NSCLC of different phenotypes (72 ACC, 33 SCC, 64 PD-L1 positive, 41 PD-L1 negative) were analysed. Various segmentation techniques were employed. Radiomic features were obtained from the original and transformed images using the PyRadiomics package. Records were split into training and test sets in the ratio of 7:3. Feature reduction involved the Mann-Whitney U test, LASSO regression, and Spearman correlation analysis. A logistic regression model was developed using the selected features, and performance was assessed with ROC curve, AUC score, and other metrics.

The optimal model achieved an AUC of 0.783 (95% CI: 0.625, 942), with high accuracy (81.25%), sensitivity (90.00%), PPV (81.81%), and NPV (80.00%).

Radiomic features derived from 18 F-FDG-PET/CT images can potentially differentiate between PD-L1 positive and negative NSCLC. Consequently, radiomics with multimodal imaging presents a promising non-invasive approach for selecting patients who may benefit from targeted immunotherapy.