Real-time AI-assisted quality control during nasopharyngolaryngoscopy: a randomized controlled trial

Nasopharyngolaryngoscopy (NPL) is widely used to examine the nasopharyngolaryngeal anatomical sites. The quality of NPL depends on the endoscopist's performance, and incomplete examinations may contribute to missed findings in practice. Here, we developed ENDOVISTA-ENT, an intelligent quality control system trained on NPL videos from 3,630 patients. The system can monitor anatomical coverage in real time during NPL procedures. It is not designed to detect lesions. By integrating into the existing NPL workflow, it provides endoscopists with real-time feedback on anatomical coverage, examination progress, and procedure duration. To evaluate its effect, we conducted a prospective, double-centre, randomized controlled trial registered in the Chinese Clinical Trial Registry (ChiCTR2400091245). A total of 318 patients were randomly assigned to undergo ENDOVISTA-ENT-assisted or conventional NPL examination. The primary outcome was coverage of predefined anatomical sites. Results showed that ENDOVISTA-ENT-assisted NPL examinations achieved signi6cantly higher mean anatomical coverage than conventional examinations (93.08% vs. 83.50%, P < 0.0001). Importantly, this improvement occurred without significantly increasing examination time. Subgroup analyses revealed benefits across all experience levels, particularly among junior endoscopists. These findings suggest that a real-time AI-assisted quality control system can support a more standardized NPL workflow and improve endoscopists' procedural completeness during NPL.