Utilidad Diagnóstica y Pronóstica de la Radiogenómica Basada en Inteligencia Artificial en Imágenes PET/CT y MRI para la Caracterización Tumoral en Cánceres Sólidos: Revisión Sistemática de Estudios Recientes

This systematic review aimed to evaluate the diagnostic and prognostic utility of artificial intelligence–based radiogenomics applied to PET/CT, MRI, and PET/MRI imaging in solid cancers. A comprehensive search of major databases was conducted for studies published between 2020 and 2025, following PRISMA 2020 guidelines. Twenty original studies using machine learning or deep learning models integrating radiomic features with genomic or transcriptomic data were included, most of them retrospective multicenter cohorts. Radiogenomic models demonstrated high diagnostic performance, with AUC values ranging from 0.80 to 0.91, particularly for predicting key molecular alterations such as IDH, MGMT, EGFR, and KRAS. Prognostic models showed consistent ability to estimate overall survival, progression-free survival, and treatment response across gliomas, lung cancer, breast cancer, and prostate cancer. Multimodal PET/MRI deep learning approaches achieved the highest performance, outperforming traditional clinical and imaging-only models. However, important methodological limitations were identified, including lack of external validation, heterogeneity in segmentation and feature extraction, and reliance on retrospective designs. In conclusion, AI-based radiogenomics shows strong potential for improving tumor characterization and outcome prediction in solid cancers, but prospective validation and methodological standardization are required before clinical implementation.