AI-Enhanced Radiomics for Early Risk Stratification of Neuroblastoma in East Mediterranean Pediatric Patients: A Multi-Center Study
Giovanni Romano¹, Renata Bianchi², Lefèvre Sylvain³
Keywords:
neuroblastoma, pediatric oncology, AI radiomics, early diagnosis, East MediterraneanAbstract
Neuroblastoma accounts for approximately 8–10% of pediatric cancers in the East Mediterranean region, with delayed diagnosis contributing to poor outcomes. This multicenter prospective study recruited 476 newly diagnosed neuroblastoma patients (ages 0–14) from seven oncology centers between January 2022 and December 2024. The aim was to evaluate the performance of artificial intelligence (AI)-driven radiomics in predicting high-risk tumor profiles using initial MRI scans.
Baseline demographic, clinical, and imaging data were collected. A total of 32 radiomic features were extracted per tumor, including texture, shape, and intensity metrics. High-risk disease was defined based on INRG staging, MYCN amplification, and metastatic status. Of the cohort, 182 children (38.2%) were classified as high-risk.
The AI model, based on a convolutional neural network trained on 70% of the cohort and tested on 30%, demonstrated strong discriminatory power. Key predictors included entropy score > 5.8, sphericity < 0.65, and non-uniformity in the gray-level co-occurrence matrix. The final model achieved an AUC of 0.92 (95% CI: 0.89–0.95), with a sensitivity of 85.1% and a specificity of 88.3%. Notably, the model correctly predicted 91% of MYCN-amplified cases without requiring biopsy or genetic profiling.
Patients flagged as high-risk by the model were referred earlier for intensive chemotherapy, with a 22% reduction in treatment delay compared to the standard diagnostic workflow (p < 0.001). Follow-up analysis revealed that early therapeutic decision-making based on AI predictions resulted in improved interim response rates (CR or PR in 76.4% of patients vs. 61.1% of controls, p = 0.02).
This study provides strong evidence for integrating AI-enhanced imaging analytics into pediatric oncology workflows in low- to middle-income settings, enabling earlier intervention and improving survival potential in high-risk neuroblastoma cases.
