[Deep learning-based endoscopic diagnosis of nasopharyngeal carcinoma: model development and cloud deployment]

Objective: To develop a deep learning-assisted diagnostic model based on white light imaging (WLI) and narrow band imaging (NBI) endoscopic images for nasopharyngeal carcinoma (NPC), and to further explore its potential clinical application in remote diagnostic systems. Methods: Endoscopic image data were retrospectively collected from 1 262 subjects who underwent nasopharyngeal screening at the Department of Otolaryngology, The First Affiliated Hospital of Sun Yat-sen University between July 2018 and September 2021. A total of 9 370 WLI and 5 558 NBI images were included. NPC recognition models were developed and validated separately for WLI and NBI using the InceptionResNetV2 deep neural network architecture. Results: An artificial intelligence-based diagnostic model for NPC (NPC-CAD) was successfully established, achieving diagnostic accuracies of 96.03% and 91.16% in the WLI and NBI test sets, with AUC of 0.99 and 0.96 respectively. A cloud-based intelligent diagnostic platform was subsequently developed and demonstrated preliminary feasibility and performance in remote healthcare scenarios. Conclusion: This study presents a novel AI-assisted diagnostic model for NPC that integrates deep learning with nasal endoscopic image analysis and introduces a prototype cloud-based diagnostic platform. The model shows strong potential to improve early screening efficiency and to reduce missed diagnoses, particularly in resource-limited settings.