Convolutional Neural Network to Assess Phayngeal and Laryngeal Pathology and Function on Nasopharyngolaryngoscopy

Worldwide, 686,000 new head and neck (H&N) cancers are diagnosed yearly, and 375,000 people will die annually. Human papillomavirus (HPV) is responsible for an increasing subset of H&N malignancies called oropharyngeal squamous cell carcinomas (OPSCC). Although it has a better prognosis than HPV-negative cancers, the 5-year overall survival is 80%; hence, new methods are needed to improve the survival of this deadly cancer.

The Need

Current methods for diagnosing OPSCC include PET, endoscopy, and subsequent biopsy of suspected lesions. Unfortunately, it's challenging to differentiate malignant lesions from healthy tissues, limiting the early detection of the disease, which is essential for eliminating cancer before it enters an advanced and untreatable stage. New diagnostic methods are needed to improve the accuracy of the diagnosis of OPSCC.

The Technology

This machine learning (ML) approach for diagnosing OPSCC uses video nasopharynx laryngoscopy (NPL) to evaluate malignant lesions. The neural network within the ML system is trained using existing annotated NPL data classified as cancer versus health tissues. The inventors have created a prototype of the system and tested it on over 80 patients undergoing treatment or follow-up care for OPSCC. Resultant data indicate the ML-based analysis tool accurately predicts active cancer versus normal tissues.

Commercial Applications

This invention can improve the diagnosis and facilitate early treatment of OPSCC and potentially other cancers.


Based on data generated by the inventors, this technology offers improvements in distinguishing OPSCC lesions versus health tissues using traditional NPL. As a result, it may help identify cancers early while eliminating unnecessary surgery in those without the disease.

Loading icon