JKAOH Journal of Korean Academy of Oral Health

ISSN(Print) 1225-388X ISSN(Online) 2093-7784


Home All Articles View

Original Article

Journal of Korean Academy of Oral Health 2019; 43(4): 210-216

Published online December 30, 2019 https://doi.org/10.11149/jkaoh.2019.43.4.210

Copyright © Journal of Korean Academy of Oral Health.

Possibility of predicting missing teeth using deep learning: a pilot study

Seon-Jip Kim1, Dohyoung Rim2, Jeong Uk Heo3, Hyun-Jae Cho1

1Department of Preventive Dentistry and Public Oral Health, School of Dentistry, Seoul National University, 2Department of Cognitive Science, Yonsei University, Seoul, 3Goodwill Dental Hospital at Hadan, Busan, Korea

Received: November 23, 2019; Revised: December 3, 2019; Accepted: December 4, 2019


Objectives: The primary objective of this study was to determine if the number of missing teeth could be predicted by oral disease pathogens, and the secondary objective was to assess whether deep learning is a better way of predicting the number of missing teeth than multivariable linear regression (MLR).
Methods: Data were collected through review of patient's initial medical records. A total of 960 participants were cross-sectionally surveyed. MLR analysis was performed to assess the relationship between the number of missing teeth and the results of real-time PCR assay (done for quantification of 11 oral disease pathogens). A convolutional neural network (CNN) was used as the deep learning model and compared with MLR models. Each model was performed five times to generate an average accuracy rate and mean square error (MSE). The accuracy of predicting the number of missing teeth was evaluated and compared between the CNN and MLR methods.
Results: Model 1 had the demographic information necessary for the prediction of periodontal diseases in addition to the red and the orange complex bacteria that are highly predominant in oral diseases. The accuracy of the convolutional neural network in this model was 65.0%. However, applying Model 4, which added yellow complex bacteria to the total bacterial load, increased the expected extractions of dental caries to 70.2%.
On the other hand, the accuracy of the MLR was about 50.0% in all models. The mean square error of the CNN was considerably smaller than that of the MLR, resulting in better predictability.
Conclusions: Oral disease pathogens can be used as a predictor of missing teeth and deep learning can be a more accurate analysis method to predict the number of missing teeth as compared to MLR.

Keywords: Deep learning, Linear regression, Missing teeth, Real-time PCR, Periodontitis