Development and Validation of a Deep Learning System for Nasopharyngeal Carcinoma Using Endoscopic Images

Development and Validation of a Deep Learning System for Nasopharyngeal Carcinoma Using Endoscopic Images: a Multi-center Prospective Study

Develop a deep learning algorithm via nasal endoscopic images from eight NPC treatment centerto detect and screen nasopharyngeal carcinoma(NPC).

Nasopharyngeal carcinoma (NPC) is an epithelial cancer derived from nasopharyngeal mucosa. Nasal endoscopy is the conventional examination for NPC screening. It is a major challenge for inexperienced endoscopists to accurately distinguish NPC and other benign dieseases. In this study, we collcet multi-center endoscopic images and train a deep learning model to detect NPC and indicate tumor location. Then, the model perfomance will be compared with endoscopists and be tested prospectively with external dataset.

Inclusion Criteria:

• The quality of endoscopic images should clinical acceptable.
• Patients were diagnosed with biopsy(NPC, benign hyperplasia). Control corhort(normal nasopharynx) don't require bispsy result.

Exclusion Criteria:

• images with spots from lens flares or stains, and overexposure were excluded from further analysis.
• image can not expose most part of lesion clearly.