Diagnostic performance of AI-assisted intraoperative imaging for parathyroid identification: a systematic review.

BACKGROUND: Intra-operative identification of parathyroid glands might be technically challenging and is a major source of hypocalcemia and nerve injury. Artificial-intelligence (AI)-based computer-vision systems have emerged as a potential adjunct. METHODS: A comprehensive search strategy was performed to identify studies using any machine-learning or deep-learning model for intraoperative parathyroid localization or viability assessment in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The protocol on PROSPERO: CRD420251104234. RESULTS: From 1124 records, 12 studies met inclusion criteria. Among the included studies, nine focused on localization of abnormal glands, while others addressed normal gland identification or perfusion assessment. AI systems utilized various imaging modalities including near infrared autofluorescence, red-blue-green surgical video, and indocyanine green angiography. These models have achieved high diagnostic accuracy, with most systems demonstrating >85% sensitivity and specificity for parathyroid identification. CONCLUSIONS: AI-driven systems demonstrate promising diagnostic performance for intraoperative parathyroid identification across multiple imaging modalities. Although results are encouraging, multicenter datasets, surgeon trust and workflow integration, and trials linking AI guidance to clinical outcomes are essential before routine adoption.