AI Enhances NBI Laryngeal Cancer Diagnosis with High Accuracy

A novel single-view contrastive learning model achieved 96.12% accuracy in classifying laryngeal leukoplakia from NBI images. The method augments small datasets using patch-based sampling and pseudo-labeling, enabling fine-grained classification across six laryngeal tissue types. It outperforms existing models in sensitivity and specificity, enhancing computer-aided diagnosis of laryngeal cancer. This advancement supports early, accurate diagnosis using narrow-band imaging (NBI) under limited data conditions.