Applications of machine learning (ML) in endocrinology have expanded rapidly over the past two decades, with thyroid-related research dominating the field, according to a comprehensive explanatory review published in the journal Endocrine.
The authors, led by Alisha Hubalewska-Dydijk, MD, chair of the Department of Endocrinology, Jagiellonian University School of Medicine in Krakow, Poland, identified 1,130 original studies applying ML techniques to nondiabetic endocrine diseases published from January 2000 to December 2024, highlighting their increasing use in imaging, risk prediction, and treatment response modeling, as well as persistent limitations. in validation and clinical implementation.
Researchers used PubMed to search for original full-text studies in English using ML techniques in thyroid, pituitary, adrenal, and parathyroid diseases. They excluded studies focused on diabetology because of extensive prior coverage. Included research spanned image-based ML, radiomics, electronic medical record analysis, and molecular and omics-based modeling.
Of the 1,130 studies analyzed, the majority were related to thyroid disease (68%), followed by pituitary disease (20%), adrenal gland disease (7%), and parathyroid disease (5%). Most studies were retrospective and single-center.
thyroid disease
ML was most frequently applied for ultrasound-based evaluation of thyroid nodules. Several deep learning models have demonstrated diagnostic performance comparable to or exceeding that of expert radiologists. In a multicenter study cited in the review, a deep learning system reduced unnecessary fine-needle aspiration biopsies by 27% while maintaining diagnostic accuracy. The ML model also showed high performance in molecular risk stratification, including predicting malignancy, detecting lymph node metastasis, and predicting BRAFV600E mutations. In cytology, we combined refractive index and ML analysis of stained images to achieve up to 100% accuracy in differentiating between benign and malignant samples in a small study.
Disorders of the pituitary gland
For pituitary imaging, ML-based radiomics differentiated cystic pituitary adenomas from Rathke’s cleft cysts with an area under the curve (AUC) of 0.848. A texture-based ML model predicted response to first-generation somatostatin receptor ligands in acromegaly with an AUC value of 0.847. Postoperative outcome prediction models such as remission, hypopituitarism, hyponatremia, and diabetes insipidus have also been reported, mainly using preoperative imaging and clinical variables.
adrenal dysfunction
Although small in number, adrenal ML studies have shown high diagnostic performance in certain applications. Radiomics-based ML differentiated lipid-poor adenomas from malignant lesions and pheochromocytomas, with some models achieving AUC values greater than 0.94. In primary hyperaldosteronism, ML-based clinical scoring systems have a sensitivity of >90% and can reduce unnecessary screening tests by up to 32.7% without missing surgically curable cases. By combining steroid profiling and ML, we achieved a well-balanced classification of adrenal tumor subtypes with a high accuracy of 97%.
parathyroid disease
The application of ML in parathyroid diseases focuses on improving detection and surgical outcomes. Deep learning applied to fluorocholine PET/CT identified hyperactive parathyroid tissue with 83% detection accuracy. Intraoperative ML-assisted imaging techniques achieved up to 100% sensitivity and >90% specificity for parathyroid gland identification. A random forest model predicting postoperative hypocalcemia after thyroidectomy reached an AUC of 0.928 in the validation cohort.
Restrictions
The authors noted recurring limitations across subspecialties, including lack of model transparency, data imbalance, small sample sizes, and heavy reliance on retrospective designs. There was little external validation or standardized reporting. The focus of research on thyroid diseases highlights research disparities and may limit advances with ML in rarer endocrine diseases.
“High-quality, well-designed ML-based research is needed in endocrinology, as in other medical fields,” the authors noted. “The validation and integration of ML models into clinical practice requires expert attention and careful oversight. Interdisciplinary collaboration between medical professionals, data scientists, and AI experts is essential to realize the full potential of this technology. Thus, despite the challenges, ML technology has the potential to bring significant benefits to the endocrine field.”
The authors reported no relevant discrepancies.
