Videomics and artificial intelligence in endoscopic diagnosis of laryngeal lesions: mapping current evidence through a scoping review

Laryngeal lesions are common and despite advances like high-definition videolaryngoscopy and enhanced imaging modalities such as narrow-band imaging, laryngoscopy remains operator-dependent. In this setting, artificial intelligence (AI) represents a promising tool to support clinical evaluation. This scoping review evaluated the current applications of AI in the endoscopic diagnosis of laryngeal lesions. A comprehensive search of MEDLINE and Scopus databases included 35 studies addressing AI-based detection, classification, or segmentation of laryngeal pathologies. Detection models frequently achieved real-time inference speeds and strong performance metrics, although external validation was limited. Classification studies showed particularly robust results for binary tasks distinguishing high-risk from low-risk lesions, with some models achieving sensitivity and accuracy exceeding 90%. Segmentation models demonstrated the potential for precise delineation of cancer margins, a capability of notable relevance for surgical planning and intraoperative decision-making. Despite promising advances, heterogeneity in study design, limited external validation, and reliance on single-centre datasets currently restrict broad clinical implementation. Nonetheless, the emerging integration of AI into laryngeal endoscopy represents a significant step toward reproducible and accessible diagnostic assessment.