Machine learning-based classification of adrenal tumors using clinical, hormonal, and body composition data.

Area of Science:

• Endocrinology
• Oncology
• Artificial Intelligence

Background:

• Accurate diagnosis of diverse adrenal tumors is challenging.
• Subtypes include mild autonomous cortisol secretion (MACS), adrenal Cushing's syndrome (ACS), primary aldosteronism (PA), pheochromocytoma (PCC), and nonfunctioning adrenal adenomas (NFAs).
• Current diagnostic methods can be complex and invasive.

Purpose of the Study:

• To develop a machine learning (ML)-based single-step diagnostic method for adrenal tumors.
• To integrate clinical data, serum adrenal hormone profiles (SAPs), and body composition data for improved differentiation.
• To enhance diagnostic accuracy and reduce the need for invasive procedures.

Main Methods:

• Developed ML models using data from 641 patients with adrenal tumors.
• Integrated 32 clinical data points, 49 SAP markers, and 15 body composition data points.
• Randomly divided patients into training (4:1 ratio) and test cohorts.

Main Results:

• The best ML model achieved a balanced accuracy of 0.78 and an AUC of 0.89 for differentiating all five tumor types.
• High accuracies were observed when distinguishing specific tumors from nonfunctioning adrenal adenomas (NFAs), with AUCs up to 0.99.
• Serum adrenal hormone profiles (SAPs) were the most critical features; body composition data had minimal contribution.

Conclusions:

• The ML model demonstrates high diagnostic accuracy for various adrenal tumor subtypes.
• Integration of clinical, body composition, and SAP data aids in differentiating adrenal tumors.
• This AI-driven approach can potentially improve clinical decision-making and reduce invasive testing.