Detecting Dental Caries Through Captured Images Using The Machine Learning Technology Teachable Machine

Abstract

Introduction: The aim is to apply artificial intelligence to identify deep tooth decay using the open-source tool Teachable Machine. Material and Methods: The study was conducted on 2063 digital images, including 1563 images with deep tooth decay and 500 images without deep tooth decay. Results: Out of the total 1563 images with deep tooth decay, when using the recognition tool, 1512 images were correctly identified (96%), and 51 images went undetected, accounting for 4%. Out of the total 2063 images, including both images with and without deep tooth decay, 1512 images were correctly identified (73.3%), and 551 images (26.7%) were not detected to have tooth decay. Conclusion: Through the study on 1563 images with deep tooth decay using the Teachable Machine learning tool, the results were promising with a high accuracy rate of 96%. However, on the mixed dataset of 2063 images, the accuracy rate for identifying images with tooth decay was only 73.3%. The difference is attributed to the early appearance of tooth decay, as its color closely correlates with normal tooth enamel. Therefore, the research team suggests the need for more data on this type of decay to enable more accurate classification and identification.