Oro Vision: Deep Learning for Classifying Orofacial Diseases

This experiment is an attempt to apply deep learning techniques to orofacial image analysis. Health promotion is recognized as a viable approach to preventing diseases and disorders and promoting changes in health behaviors or practices. Each year, oral cancer kills more people in the US than does cervical cancer, malignant melanoma, or Hodgkin's disease. A first line of defense against oral diseases is an orofacial selfexamination. The goal of this experiment titled "Oro Vision" is to provide an assessment tool for field workers to perform initial examinations of orofacial diseases, using a camera enabled mobile phone. For this experiment, we chose to implement Oro Vision to detect mouth sores. The goal is to extend this model to identify several other Oral diseases such as Thrush, Leukoplakia, Lichenplanus, etc. One variety of mouth sore, referred to as the "cold sore" is highly contagious and an infected person can easily pass on the infection to another person just through skin to skin contact. "Oro Vision" is implemented as an HTML5 mobile responsive web app that can be accessed through any mobile or standard browser. Oro Vision uses deep learning to train a model and subsequently uses this trained model to distinguish a cold sore from a canker sore. In addition, an accurate diagnosis by a trained healthcare professional is required before any kind of treatment is discussed since several other conditions of the mouth including oral cancer may mimic canker sores.