Machine learning in dentistry

• Main Author: Yuen, Priscilla Li Jie
• Other Authors: Mohammed Yakoob Siyal
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/166817
• Institution: Nanyang Technological University

With the birth of Artificial Intelligence (AI) in the 1950s, Machine Learning (ML) came along, and then Deep Learning (DL). The idea of AI is to have human intelligence defined in a way such that a machine can imitate and execute tasks as humans would. ML, a subset of AI, uses algorithms that analyse large amounts of statistical data to build models that are able to automatically learn and improve. DL is a subset of ML that is inspired by the way a human brain filters information using neurons. With the improvements in hardware like Graphical Processing Unit (GPU), DL has allowed advancements in a machine’s ability to conduct tasks like speech recognition, visual recognition, and language understanding [1,2]. Visual recognition is a huge feat for machines as they do not process images the way human eyes view and recognise a scene. This subset of DL, Computer Vision (CV), is a major tool in various industries like automobile industry for self-driving cars, security industry for facial recognition, agriculture industry for analysis of grain quality, and even the medical industry for automatic diagnostic imaging. [3] Dentists use X-rays to diagnose oral abnormalities. [4] Panoramic X-rays are commonly used as they capture the entire oral cavity which allows for a thorough examination. In recent years, DL models have been trained using CV techniques to perform tasks that can assist the dentists in analysing these medical images – providing initial reports of X-rays or second opinions that highlight specific abnormalities like cavities. DL models, however, require large training data to produce accurate results. Public datasets of dental imaging are also scarce due to privacy issues and more critically, its requirement for industry experts to label each image in the dataset makes it even more inaccessible. With that, this project aims to implement a comprehensive system using the recently published Tufts Dental Database, consisting of pre-processing methods and a semantic segmentation model to produce dense pixel-wise segmentation maps of dental radiographs, in hopes of aiding dentists with their diagnoses. In this thesis, the Literature Review section will evaluate the existing deep learning models, image pre- 7 processing methods and segmentation models. The models selected and the results obtained will be presented next in the Methods & Results section. Then, a Web Application segment will present a web application created that integrates the final model with a user interface. Lastly, a conclusion will be made with a roundup of the findings from this study along with potential areas for the future studies.