Nasopharyngeal carcinoma (NPC) is a highly aggressive malignancy, with most patients presenting at locally advanced stages. While immune checkpoint inhibitors, such as PD-1 blockade, have reshaped treatment strategies, only a minority of patients achieve durable benefit. Accurate biomarkers for predicting treatment response remain an urgent unmet need.

A multicenter study led by Prof. Shuixing Zhang and Prof. Bin Zhang from the First Affiliated Hospital of Jinan University enrolled 246 patients with locally advanced NPC treated with immunotherapy. By applying artificial intelligence algorithms, the team extracted and selected optimal radiomic features from medical imaging to construct a predictive model(Fig. 1). Results demonstrated that this AI-based radiomics model achieved an AUC of 0.760, significantly outperforming traditional clinical models (AUC 0.559) in predicting treatment response. For prognosis, the optimal model reached a C-index of 0.858, accurately stratifying patients into high- and low-risk groups.

Beyond predictive performance, the study also explored the biological interpretability of the model. Through image–pathology correlation analysis using whole-slide H&E and IHC images, researchers uncovered strong associations between radiomic features and key immune cell markers, including CD45RO, CD8, PD-L1, and CD163. These findings reveal a clear link between imaging-derived features and the immune landscape of the tumor microenvironment, providing biological validation of the radiomics approach.

Together, this work highlights the promise of radiomics as a powerful, non-invasive tool for precision immunotherapy in NPC. By combining advanced imaging analytics with pathology correlation, the study not only improves predictive accuracy but also bridges radiomic signatures with tumor biology, offering new insights into patient stratification and personalized treatment.

The complete study is accessible via DOI: 10.34133/research.0749