Automated Dental Cavity Detection System Using Deep Learning and Explainable AI.

Researchers

Journal

Modalities

Models

Abstract

Impacting over 3.9 billion people, dental cavities requires a trained dentist for diagnosis. Unfortunately, barriers such as dentophobia, limited dentist availability, and lack of dental insurance prevent millions from receiving care. To address this, an Artificial Intelligence system was developed that detects cavity presence on photographs and visually explains the rationale behind each diagnosis. While previous systems only detected cavities on one extracted tooth showing one tooth surface, this study’s system detects cavities on photographs showing multiple teeth and four tooth surfaces. For training, 506 de-identified images from online sources and consenting human participants were collected. Using curriculum learning, a ResNet-27 architecture proved to be most optimal after achieving 82.8% accuracy and 1.0 in sensitivity. Visual explanations for the system’s diagnoses were also generated using Local Interpretable Model Agnostic Explanation. This system can explain its diagnoses to users in an understandable manner, which is a crucial skill employed by dentists.©2022 AMIA – All rights reserved.