Answers are grounded in this paper and cite their sources.
OBJECTIVE(S): To develop and validate an artificial intelligence (AI) system for standardized Ni classification of nasopharyngeal narrow-band imaging (NBI) and to evaluate whether AI assistance improves physician diagnostic consistency and biopsy-related decision making in a reader-study setting. METHODS: In this multicenter diagnostic study, 5404 high-quality nasopharyngeal NBI images were retrospectively collected. Lesion segmentation and five-class Ni classification networks were trained independently. Performance was evaluated in an internal test set, two independent external cohorts, and a recorded-video validation set. A prospective randomized multi-reader crossover study involving 12 physicians and 200 images from CHSZCAMS assessed the effect of AI assistance on diagnostic performance, inter-observer agreement, and potentially unnecessary biopsy decisions. RESULTS: In the two external cohorts, the model achieved accuracies of 0.889 and 0.873, with AUCs of 0.976 and 0.967, respectively. In recorded-video validation, accuracy was 0.888 and AUC was 0.974. AI assistance improved Fleiss kappa from 0.610 to 0.788 in junior physicians and from 0.744 to 0.870 in senior physicians (both p < 0.05). Potentially unnecessary biopsy decisions decreased from 21.6% to 10.3% in junior physicians and from 13.7% to 8.0% in senior physicians. CONCLUSION: AI-assisted Ni classification showed consistent diagnostic performance across validation cohorts and improved physician agreement across experience levels. In this reader-study setting, AI assistance reduced potentially unnecessary biopsy decisions, supporting further evaluation as a second-reader tool for nasopharyngeal NBI interpretation.
No takes yet. Share an insight, caveat, or question.
Huang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: