Development of a computed tomography radiomics and CD38 integrated model: predicting immunotherapy response and investigating biological implications in non-small cell lung cancer.

Background: Non-small cell lung cancer (NSCLC) accounts for 80-85% of lung cancers and remains the leading cause of cancer-related mortality worldwide. Immune checkpoint inhibitors (ICIs) improve survival in selected patients, yet their clinical utility is limited by low objective response rates, challenges in patient selection, and suboptimal performance of existing biomarkers. Radiomics can quantitatively characterize tumor heterogeneity on computed tomography (CT) imaging and shows promise in predicting treatment response and immune microenvironment profiles. CD38, a key immunosuppressive molecule, impairs T-cell function and recruits suppressive cells via the adenosine pathway. This study aimed to integrate radiomic features with CD38 expression to build a multimodal prediction model and investigate its underlying biology. Methods: We retrospectively included 45 NSCLC patients receiving ICIs (training cohort: n=31; validation cohort: n=14). A total of 1,223 CT radiomic features were extracted and selected by least absolute shrinkage and selection operator (LASSO) regression to construct a radiomics model. CD38 expression was quantified by immunohistochemistry and combined with radiomics in a fusion model. An independent validation cohort (n=89) was derived from The Cancer Imaging Archive (TCIA) and Gene Expression Omnibus (GEO) datasets. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Overall survival (OS) differences were assessed by Kaplan-Meier analysis. Molecular mechanisms were explored using differential gene expression, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment, Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT), and single-sample gene set enrichment analysis (ssGSEA). Results: The radiomics model achieved area under the curves (AUCs) of 0.734 and 0.878 in the training and validation cohorts, respectively. Incorporating CD38 increased the AUC in the training cohort to 0.801, with significant improvement in NRI (0.952, P<0.001) and IDI (0.148, P=0.001). A nomogram integrating clinical variables enabled individualized prediction. High-response patients had significantly longer OS than low-response patients [hazard ratio (HR) =0.45, P=0.001], with consistent advantages in subgroups such as age >64 years and T1-2 stage. Mechanistically, high responders exhibited enrichment of epithelial differentiation genes and T-cell activation pathways, with increased CD8 Conclusions: Integrating radiomic features with CD38 expression significantly enhances specificity and clinical applicability in predicting immunotherapy response, while elucidating molecular and immunological mechanisms underlying response heterogeneity. The model supports individualized risk stratification and precision therapy planning; however, larger multicenter prospective studies and multi-omics integration are warranted to further validate and optimize its utility.