Thesis
Prediction of canine eruption problems and other developmental anomalies in panoramic radiographs using machine learning
University of Iowa
Master of Science (MS), University of Iowa
Spring 2021
DOI: 10.17077/etd.006091
Abstract
IntroductionEruption problems and developmental dental anomalies are common clinical complications encountered by orthodontists, pediatric dentists, and general dentists. In most instances, early identification of these problems allows for proper intervention and improved outcomes for patients. Over the past decade, the interest in implementation of machine learning methods in orthodontic diagnosis and treatment planning as increased exponentially. The main goal of our study is to evaluate the performance of deep learning models in predicting maxillary canine impactions and other developmental dental anomalies from panoramic X-rays. We perform the evaluation of these models following two specific aims. The first aim was to evaluate the overall performance of several different deep learning models with our panoramic radiograph image set to select the best model for our study. The second aim was to adjust the parameters of the chosen model in an effort to improve its overall performance and also to test its performance under a specific subset of mixed dentition images derived from our main panoramic image set.
Materials and Methods
Our study consisted of two panoramic radiograph image sets, an impaction/anomaly group (n=608) and a control/normal group (n=1356). Several neural network models (ResNet-18, VGG11, ResNet-50, VGG16, Inception v2, and Inception v3) were compared in their performance for this study. The ResNet-18 convolutional neural network model was selected and was trained, verified, and tested in an unsupervised approach with both image sets. Several parameters of the model were modified, and performance data were analyzed and interpreted. The modified model was used a second time to analyze its performance on a subgroup set of images composed of isolated mixed dentition radiographs from our initial image set.
Results
The ResNet-18 model was selected based on its superior initial performance. For our complete panoramic radiographs established for this study, the ResNet-18 model’s performance metrics ranged from 0.38 to 0.62, while the F1 score ranged from 0.38 to 0.47, and accuracy values ranged from 0.49 to 0.53. The isolated mixed dentition image set showed a wider range of performance values (0.21 to 0.85), with F1 score ranging from 0.28 to 0.45 and accuracy values ranging from 0.33 to 0.52. The true positive rate was generally higher than the true negative rate.
Conclusions
Following the classification of our panoramic radiographs, we identified the ResNet-18 model as the superior neural network amongst the models considered. The ResNet-18 neural network model used on our complete panoramic radiographs showed performance values ranging from 0.38 to 0.62. The model for the isolated mixed dentition image set showed more variable performance and slightly poorer performance compared to the total data set. The model generally performed better in classifying/identifying images where impaction was already present compared to predicting future impaction. The model showed better performance in later stages of development. This study is a promising foundation for future studies of machine learning in prediction of eruption problems and dental anomalies.
Details
- Title: Subtitle
- Prediction of canine eruption problems and other developmental anomalies in panoramic radiographs using machine learning
- Creators
- John Welk
- Contributors
- Shankar Rengasamy Venugopalan (Advisor)Lina M Moreno-Uribe (Committee Member)Kyungsup Shin (Committee Member)Brian J Howe (Committee Member)Michael A Callan (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Orthodontics
- Date degree season
- Spring 2021
- DOI
- 10.17077/etd.006091
- Publisher
- University of Iowa
- Number of pages
- ix, 58 pages
- Copyright
- Copyright 2021 John Welk
- Language
- English
- Description illustrations
- illustrations (some color)
- Description bibliographic
- Includes bibliographical references (pages 46-58).
- Public Abstract (ETD)
Eruption problems and developmental dental anomalies are relatively common problems encountered by orthodontists, pediatric dentists, and general dentists. In most instances, early identification of these eruption problems allows for proper treatment and improved results for patients. Over the last few decades, there has been increasing interest in using computer programs to help physicians and dentists improve the accuracy and efficiency of their decisions and improve the quality of care for their patients.
Our study used 1356 panoramic radiographs from patients that had normal dental development and 608 panoramic radiographs from patients that had impacted maxillary canines. We used a specialized computer program to try to predict whether maxillary canines would become impacted by only using a panoramic x-ray image. We measured how well the computer program performed in predicting these images.
Our computer program’s performance ranged from 38-62% in predicting whether or not there was going to be impacted maxillary canines for a particular patient based on their panoramic radiograph. These performance scores dropped for our second test with x-rays from subjects who were earlier in their dental development. Our program tended to predict more images as impacted than there truly were.
We were able to identify the best computer program amongst those considered for use in our study. The computer’s performance was better for x-rays which were taken later in dental development compared to earlier on in development. This study is a promising foundation for future studies.
- Academic Unit
- Orthodontics; Craniofacial Anomalies Research Center
- Record Identifier
- 9984096976502771
Metrics
25 File views/ downloads
179 Record Views