Artificial intelligence in pheochromocytoma and paraganglioma: from diagnostic modeling to clinically integrated...

Over the past decade, interest in artificial intelligence (AI) has expanded rapidly in medicine. In pheochromocytoma and paraganglioma (PPGL), AI-based studies have increased but remain constrained by limited data typical of rare diseases. This review provides current AI and machine learning (ML) applications in PPGL across diagnostics, imaging, molecular classification, perioperative risk prediction, and disease stratification and outlines practical pathways for clinically robust translation. More specifically, AI methods have been applied to improve interpretation of catecholamine metabolites by moving from binary thresholds toward probabilistic outputs and integrated clinicobiochemical models. Omics-based ML approaches (metabolomics, methylomics, and transcriptomics) and radiomics models show promising discriminatory performance for disease detection, tumor characterization, and risk stratification, including predictors of aggressive behavior and perioperative instability. However, most studies are retrospective, single-center, and modestly sized and frequently lack independent external validation. Key translational barriers include heterogeneity of imaging/omics pipelines, insufficient reporting of calibration and clinical utility, limited interpretability, and sparse attention to governance and regulatory pathways. Although ML integration into clinical decision support systems could address several of these gaps, implementation studies in PPGL are scarce, and evidence indicates that probability scores alone may not improve expert decision-making without careful workflow integration and evaluation.