However, hurdles encountered with the latest AI iterations of chatbots underscore the need to proceed with caution.
“In contrast to other medical fields, endocrinology is not connected to a single organ structure; rather, it is a complicated biological system of hormones and metabolites, [intertwined with] various receptors, signaling pathways and intricate feedback mechanisms,” explained the authors of a recent article on the issue in Nature Reviews Endocrinology.
With interconnections that are “often beyond the comprehension and reasoning capabilities of the human brain, AI [is anticipated] to be exceptionally well-suited to tackle this remarkable heterogeneity and complexity,” they wrote.
Since the first regulatory approvals for AI-based technology were granted back in 2015, endocrinology has already been revolutionized by AI-based tools, most notably with AI biosensors for continuous glucose monitoring systems alerting patients of glucose levels, and automated insulin-delivery systems.
AI-based machine learning has also ushered in improved detection and analysis of thyroid nodules and potential malignancies, with algorithms in the analysis of radiological test images enabling detection through a deeper analysis than can be applied with individual specialists.
Likewise, the benefits of AI in imaging extend to osteoporosis.
“Imaging certainly is one of the most promising fields, including (but not limited to) conventional radiography, computed tomography, and magnetic resonance tomography,” explained Hans Peter Dimai, MD, a professor of medicine and endocrinology at the Medical University of Graz (Austria), and the past president of the Austrian Bone and Mineral Society.
“A typical indication is fracture detection, not in terms of replacing expert radiologists or orthopedists but rather in terms of supporting those who are in specialist training,” he said in an interview.
“Particularly the underdiagnosis of vertebral fractures has been an issue in past decades, with dramatic implications for the individual, since the first vertebral fracture would multiply the risk for any future fracture, and therefore requires immediate action from a physician’s side.”
The areas expected to further benefit from AI continue to grow as systems evolve, with advances being reported across a variety of endocrinologic conditions.
Papillary thyroid cancer (PTC): Central lymph node metastasis of papillary thyroid cancer is predictive of tumor recurrence and overall survival in PTC. However, few tests are able to diagnose the metastasis in the cancer with high accuracy. Using a convolutional neural network prediction model built with a deep-learning algorithm, researchers described high diagnostic sensitivity and specificity of a model, as reported in a study published in Feburary. The prediction model, developed using genetic mutations and clinicopathologic factors, showed high prediction efficacy, with validation in subclinical as well as clinical metastasis groups, suggesting broad applicability.
Adrenal tumors: Adrenal incidentalomas, or masses that are incidentally discovered when performing abdominal imaging for other reasons, can be a perplexing clinical challenge. Discovery of these is increasing as imaging technology advances. However, an AI-based machine learning approach utilizing CT is being developed to differentiate between subclinical pheochromocytoma and lipid-poor adenomas. As reported in a 2022 study, the prediction model scoring system used traditional radiological features on CT images to provide for a noninvasive method in assisting in the diagnosis and providing personalized care for people with adrenal tumors.
Osteoporosis – bone mineral density (BMD): In the diagnosis of osteoporosis, the measurement of BMD using dual-energy x-ray absorptiometry (DXA) is the gold standard. However, the availability of DXA devices in many countries is inadequate, leaving an unmet need for alternative approaches. But one AI-based algorithm shows promising diagnostic accuracy, compared with DXA, potentially providing a low-cost screening alternative for the early diagnosis of osteoporosis.
Osteoporosis – Fracture Risk Assessment Tool (FRAX): In fracture risk and prevention, the free FRAX tool, available online, is the gold standard and recommended in nearly all osteoporosis guidelines. However, several studies on AI-based tools show some benefit over FRAX, including one approach using longitudinal data with conventional spine radiographs, showing predictive accuracy that exceeds FRAX.
Osteoporosis – treatment: And for the often challenging process of treatment decision-making in osteoporosis, AI-based software, developed from more than 15,000 osteoporosis patients followed over 10 years, shows high accuracy in the prediction of response to treatment in terms of BMD increase, as described in another study. “Our results show that it is feasible to use a combination of electronic medical records–derived information to develop a machine-learning algorithm to predict a BMD response following osteoporosis treatment,” the authors reported. “This alternative approach can aid physicians to select an optimal therapeutic regimen in order to maximize a patient-specific treatment outcome.”