Application of Artificial Intelligence in Diagnosis and Management of Thyroid Disease

Artificial intelligence (AI) is transforming thyroid care, with applications spanning ultrasound risk stratification, cytopathology, histopathology, radiogenomics, prognosis, intraoperative parathyroid identification, thyroid eye disease, functional thyroid disorders, and clinical decision support. This narrative review summarizes the evidence and barriers to adoption. Ultrasound-based AI for thyroid nodule risk stratification is the most mature application. Deep learning-assisted cytopathology and intraoperative frozen-section analysis show promise in resolving indeterminate Bethesda nodules. Radiomics and digital pathology can non-invasively predict BRAF V600E, RAS, fusion alterations, and inflammatory subtypes, although most tools remain investigational. Machine learning refines recurrence prediction beyond conventional risk stratification and informs radioiodine response and dose selection. In thyroid and parathyroid surgery, AI-augmented near-infrared autofluorescence is advancing toward real-time intraoperative gland identification. AI also supports thyroid eye disease activity assessment and glucocorticoid response prediction, and emerging tools target hyper- and hypothyroidism, including differentiation of Graves' disease from thyroiditis and levothyroxine dose optimization. Large language models are improving in patient education but are not approved for diagnostic use and remain prone to fabricated citations and version drift. AI should be regarded as a decision-support adjunct that augments, rather than replaces, endocrinologist judgment, contingent on prospective multicenter validation, equitable access, transparent reporting, and defined reimbursement frameworks.