Aims: Early detection of glottic, supraglottic and hypopharyngeal carcinoma improves prognosis. Timely recognition of these malignancies is influenced by experience. Detection using Artificial Intelligence (AI), with real-time overlay during flexible laryngoscopy in the out-patient setting, may improve diagnostic performance. Methodology: This study evaluated the diagnostic performance of an unified deep learning model that extends a previously trained algorithm for glottic lesions(1) to additionally localize and classify supraglottic and hypopharyngeal lesions during flexible endoscopy. Lesion-frames extracted from endoscopy videos from two head and neck oncology centres (2012–2023) were manually annotated. The primary outcome was the performance of the unified model in lesion detection. After training, the sensitivity and positive predictive value (PPV) of this model were calculated on an independent test-set, stratified by subsite and tumor stage. Secondary, the performance of the model’s binary classification output (benign or malignant) was evaluated. Results: from 475 supraglottic and hypopharyngeal endoscopy videos, 40.059 images with a benign or malignant lesion were extracted and added to the 56.036 glottic images in the database. On the unified test-set, detection sensitivity and PPV were 73.4% (95% CI: 69.8 – 76.6) and 92.1% (95% CI: 90.9 – 93.2) respectively. Carcinoma (65% of the lesions) was correctly classified in 94.3% of cases (95% CI: 91.8 – 96.5). Conclusions: the developed model showed a promising lesion detection and an excellent cancer classification performance. This is the first study to report a real-time AI model for endoscopic detection and classification of both laryngeal and pharyngeal lesions, which might help less experienced endoscopists. 1. Wellenstein DJ, Woodburn J, et al. Detection of laryngeal carcinoma during endoscopy using artificial intelligence. Head Neck. 2023 Sep;45(9):2217-2226