Positron emission tomography-based intratumoral spatial diversity for prognosis in nasopharyngeal carcinoma

This study aimed to develop a radiomics-based framework for quantifying intratumoral heterogeneity (ITH) in nasopharyngeal carcinoma (NPC) using three-dimensional tumor subregions. The goal was to enhance conventional clinical risk stratification and investigate the biological basis of ITH.

This multicenter retrospective study included PET scans from 760 treatment-naïve NPC patients. Radiomics features were extracted from the whole tumor (WT), and the intratumoral spatial diversity score (ITSDS) was simultaneously quantified through subregion analysis. After a multi-step feature selection process, both WT and ITH models were developed. Finally, a nomogram was constructed by integrating ITSDS with clinical factors. All models were evaluated using the C-index, AUC, and calibration curves, and their performance was interpreted using SHAP. Additionally, the biological characteristics of the ITH model were analyzed using tissue proteomics.

The training, internal validation, and external validation cohorts included 480, 235, and 45 patients, respectively. The ITH model showed superior C-indexes of 0.852, 0.792, and 0.757 compared to the clinical and WT models. The nomogram demonstrated the best prognostic value, with C-indexes of 0.864, 0.817, and 0.772 in the three cohorts, respectively. The nomogram effectively classified patients into risk groups for progression-free survival (P < 0.01). Tissue proteomic analysis revealed upregulated ciliary movement, a more complex tumor microenvironment, and impaired immune function in the high-ITH group.

The nomogram integrating ITSDS biomarkers with clinical factors improved risk stratification accuracy and created a clinically translatable framework for NPC. This approach potentially offers mechanistic insights into ITH and holds promise for advancing precision oncology.