Adrenal incidentaloma refers to an adrenal tumor discovered incidentally during imaging for unrelated conditions, with computed tomography (CT) being a standard detection method in Japan. While many of these tumors are non-functioning adrenal tumors (NFAT), some are hormone-producing and require early intervention. This study aimed to construct a clinical prediction model for identifying hormone-producing adrenal tumors using CT findings, basic endocrine parameters, and clinical data.

We conducted a single-center, retrospective observational study of patients admitted for adrenal tumor evaluation at Kawasaki Medical School Hospital between April 2010 and March 2024. A multi-class random forest classification model was constructed to differentiate between NFAT, primary aldosteronism (PA), mild autonomous cortisol secretion (MACS), adrenal Cushing’s syndrome (CS), and pheochromocytoma (Pheo), using CT features, endocrine test results, and clinical characteristics.

A total of 162 patients were included: NFAT (n = 55), PA (n = 44), MACS (n = 22), CS (n = 18), and Pheo (n = 23). The model achieved an AUC of 0.911. For PA, CS, and Pheo, the precision, recall, and F1-scores ranged from 0.83 to 1.00, indicating high predictive performance. Compared to individual endocrine markers, the model showed superior diagnostic accuracy. However, MACS remained challenging to detect using this approach.

This machine learning-based clinical model demonstrated high diagnostic performance for PA, CS, and Pheo before endocrine loading tests by integrating CT imaging and clinical data. External validation using data from Tokyo Women’s Medical University is currently underway to assess generalizability across institutions.