[Development of an artificial intelligence-assisted diagnostic model for malignant head and neck tumors based on endoscopic images].

Objective: To develop an artificial intelligence model based on electronic nasopharyngolaryngoscopic images for diagnosing nasopharyngeal and laryngeal-hypopharyngeal cancers and to establish a cloud-based platform to explore its potential clinical application. Methods: A dataset comprising 23 434 endoscopic images from 3 255 subjects was retrospectively collected from five medical centers, including 15 465 laryngoscopic images and 7 969 nasopharyngoscopic images. An intelligent diagnostic model for head and neck tumors, named WSC-T, was further developed. Its performance was evaluated on both internal and external test sets for the diagnosis of laryngeal-hypopharyngeal cancer and nasopharyngeal cancer, and was compared with that of a supervised learning model without contrastive learning. A cloud-based platform was also developed to preliminarily explore its clinical feasibility. Results: WSC-T demonstrated promising diagnostic performance across all test sets. For laryngeal-hypopharyngeal cancer diagnosis, internal and external test accuracies were 94.89% and 91.04%, with AUCs of 0.98 and 0.97, respectively. For nasopharyngeal cancer diagnosis, the accuracies were 96.27% and 92.31%, with AUCs of 0.98 and 0.97, respectively. A cloud-based diagnostic platform was successfully established based on this model, enabling image uploading, automated analysis, and diagnostic result output. Conclusions: The proposed model demonstrated accurate and generalizable diagnostic performance for head and neck tumors. The cloud-based platform provides a convenient means for the practical application of the intelligent diagnostic model.