AI-BASED RADIOGENOMIC MODELS FOR PREDICTING IMMUNOTHERAPY RESPONSE IN SOLID TUMORS

Abstract

The integration of artificial intelligence with radiogenomics represents a transformative approach in precision oncology, particularly for predicting immunotherapy response in solid tumors. This study develops and validates deep learning-based radiogenomic models that combine medical imaging features with genomic biomarkers to predict patient responses to immune checkpoint inhibitors. We analyzed data from 342 patients with non-small cell lung cancer, melanoma, and renal cell carcinoma who received anti-PD-1/PD-L1 therapy between 2018 and 2022. Our convolutional neural network architecture integrated CT imaging radiomics with tumor mutational burden, PD-L1 expression, and gene expression profiles. The proposed model achieved an area under the curve of 0.87 for predicting complete or partial response, significantly outperforming clinical models (AUC = 0.64, p < 0.001). Feature importance analysis revealed that radiomic texture features combined with TMB scores contributed most substantially to prediction accuracy. This research demonstrates that AI-driven radiogenomic models can provide non-invasive, cost-effective tools for patient stratification in immunotherapy, potentially reducing treatment-related toxicities and healthcare costs while improving clinical outcomes.