Data cleansing for development of an oral cancer diagnostic support system

Abstract

Oral cancer has a mortality rate of 35.5% and has many serious impacts on QOL such as feeding, pronunciation and conditioning. But it is difficult for non-specialized doctors and dentists to detect and diagnose oral cancer at an early stage.A solution to overcome this situation is the development of a diagnostic support system using Deep Learning, which has made remarkable progress in recent years. However, at present, there is no large open data set which can be used for oral cancer analysis.Therefore, we attempted to create a data set using approximately 150,000 images taken at the Department of Oral and Maxillofacial Surgery, The University of Tokyo Hospital from April 1, 2015 to March 31, 2020. However, the image data contains unnecessary images for learning, which are taken from areas other than the oral mucosa and patient information. And it takes about 10 to 15 minutes per case for selection.Therefore, we performed these data cleansing using deep learning and automatically extracted only the images which is necessary for learning.