Radiomics-based prediction of T cell-inflamed gene expression profile and prognosis in head and neck squamous cell carcinoma.

BACKGROUND

Head and neck squamous cell carcinoma (HNSCC) has a poor prognosis, and response to immune checkpoint inhibitors is variable. The T cell-inflamed gene expression profile (GEP) predicts immunotherapy efficacy but relies on invasive methods. Radiomics offers a noninvasive alternative for integrating imaging features with GEP in HNSCC.

PURPOSE

To develop a radiomics-based model to determine the predictive value of GEP and immunotherapeutic responses on HNSCC.

METHOD

GEP scores were derived using HNSCC data in The Cancer Genome Atlas. Kaplan-Meier survival analysis, univariate landmark analysis, Cox regression, and subgroup interaction tests were used to evaluate the prognostic value of GEP for overall survival (OS) in HNSCC. Radiomic features were extracted from computed tomography images in The Cancer Imaging Archive. The prediction model was constructed on the training dataset using a gradient boosting machine (GBM). The model's predictive performance was evaluated using the receiver operating characteristic curve, calibration curve, and decision curve analysis. Radiomics scores (RS) were calculated using the GBM model to predict the probability of GEP scores and subsequently categorized into binary variables. The prognostic value of RS in HNSCC was then assessed. Additionally, we conducted gene set enrichment analysis (GSEA), immune gene expression profiling, mutation analysis, immune infiltration assessment, and immunophenoscore (IPS) evaluation to explore the molecular mechanisms underlying differences in GEP and RS.

RESULTS

A high GEP score is a protective factor for OS in HNSCC. (hazard ratio [HR], 0.692; 95% confidence interval [CI], 0.501-0.956; p = 0.026). The GBM prognostic prediction model demonstrated strong clinical utility (area under the curve [AUC]: 0.827 and 0.774 in training and validation sets). RS was higher in GEP-high patients (p < 0.001) and correlated with longer OS (p = 0.034). GSEA revealed enrichment in oxidative phosphorylation and ribosome pathways in high RS patients. Additionally, RS-high patients exhibited increased M1 macrophage infiltration and elevated IPS for both CTLA-4 combined with PD-1/PD-L1 inhibitors (p < 0.01) and PD-1/PD-L1 inhibitors alone (p < 0.05). TIGIT expression was significantly upregulated (p < 0.0001), and NOTCH1 and DNAH5 mutations were more frequent in the high RS group.

CONCLUSIONS

The radiomics-based model predicts GEP and provides prognostic insights in HNSCC. RS-high patients may benefit from immunotherapy, highlighting the potential of integrating radiomics and transcriptomics in precision oncology.