Machine learning-based classification of adrenal tumors using clinical, hormonal, and body composition data
Authors' conclusion
The ML model demonstrates high diagnostic accuracy in differentiating adrenal tumor subtypes by integrating clinical data, body composition, and SAP, potentially reducing the need for invasive procedures and aiding clinical decision-making.
Study context
- Population
- Patients with adrenal tumors, excluding adrenal metastases and adrenocortical carcinoma, Seoul National University Hospital
- Participants
- 641 patients
- Comparison
- Mild autonomous cortisol secretion (MACS) vs adrenal Cushing's syndrome (ACS) vs primary aldosteronism (PA) vs pheochromocytoma (PCC) vs NFA
- MACSn = 141
- ACSn = 64
- PAn = 265
- PCCn = 78
- NFAn = 93reference
- Outcome
- Diagnostic performance of machine learning (ML) model (clinical data, serum adrenal hormone profiles, body composition) differentiating all 5 adrenal tumor types
Methodological detailsmachine learning model development, single-centre, random 4:1 training/test split
- Machine learning model development
- Single-centre
- Random 4:1 training/test split
Performance
As reported in the abstract
Best ML model differentiating all 5 adrenal tumors
Balanced accuracy
0.78
Sensitivity
0.77
Specificity
0.93
AUC
0.89
Further findings
| Outcome | MACS | ACS | PA | PCC |
|---|---|---|---|---|
| Distinguishing each subtype from NFA | ||||
| Accuracy | 0.85 | 0.94 | 0.78 | 0.86 |
| AUC | 0.96 | 0.99 | 0.90 | 0.94 |
NFA vs other functioning adrenal tumors
- Accuracy
- 0.75
- AUC
- 0.79
Feature importance
- SAP features most critical for differentiation; body composition data contributed only minimally
Original publication
Machine learning-based classification of adrenal tumors using clinical, hormonal, and body composition data. Eur J Endocrinol. Published 29 July 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →