The application of artificial intelligence in adrenal imaging: current state of knowledge, challenges, and future directions

INTRODUCTION: Adrenal incidentalomas are common findings on abdominal imaging and require assessment of malignancy risk and hormonal activity. Although most lesions are benign adenomas, differentiating them from pheochromocytomas, adrenocortical carcinomas, metastases, and hormonally active tumors remains challenging, particularly in indeterminate cases. Artificial intelligence (AI) has emerged as a promising tool for improving adrenal lesion detection, segmentation, and characterization. METHODS: This narrative review was based on a structured literature search of PubMed, Scopus, and Web of Science for studies published between 2018 and 2025 using the terms 'AI', 'machine learning', 'deep learning', 'radiomics', 'adrenal glands', and 'adrenal imaging'. Priority was given to original studies on segmentation, detection, and lesion characterization. Computed tomography (CT) was the predominant modality, followed by magnetic resonance imaging and positron emission tomography (PET/CT). RESULTS: AI applications in adrenal imaging include gland segmentation, lesion detection, functional assessment, and lesion classification. Recent segmentation models achieved Dice coefficients approaching 0.90. Detection models reported high sensitivity and specificity, although evidence remains limited for small incidentalomas in heterogeneous real-world datasets. In lesion characterization, radiomics- and deep learning-based models showed promising diagnostic performance for differentiating hormonally active adenomas from non-functioning adenomas, lipid-poor adenomas from pheochromocytomas, benign lesions from metastases, and adenomas from adrenocortical carcinoma, with several studies reporting area under the curve (AUC) values above 0.90. Multimodal approaches integrating imaging with clinical data often outperformed imaging-only models. DISCUSSION: Despite encouraging results, most studies were retrospective, single-center studies based on small or selected cohorts, limiting reproducibility and generalizability. Variability in imaging protocols, lack of external validation, and limited workflow integration remain major barriers. Current evidence supports AI mainly as a decision support tool rather than a stand-alone diagnostic system. CONCLUSION: AI has significant potential to improve adrenal imaging by supporting faster, more accurate, and more reproducible lesion assessment. However, prospective multicenter validation, standardized methodologies, and explainable models are needed before broader clinical implementation.