Artificial intelligence for oral cancer diagnosis: What are the possibilities?

Abstract

Oral cancer could be prevented. The primary strategy is based on prevention. Most patients with oral cancer present to the hospital network with advanced staging and a low chance of cure. This condition may be related to physicians' difficulty of making an early diagnosis. With the advancement of information technology, artificial intelligence (AI) holds great promise in terms of assisting in diagnosis. Few machine learning algorithms have been developed for this purpose to date. In this paper, we will discuss the possibilities for diagnosing oral cancer using AI as a tool, as well as the implications for the population. A set of photographic images of oral lesions has been segmented, indicating not only the area of the lesion but also the class of lesion associated with it. Different neural network architectures were trained with the goal of fine segmentation (pixel by pixel), classification of image crops, and classification of whole images based on the presence or absence of a lesion. The accuracy results are acceptable, opening up possibilities not only for identifying lesions but also for classifying the pathology associated with them.