journal of prosthodontic research 65 ( 2021) 115–118

Journal of Prosthodontic Research

Original Article
A system for designing removable partial dentures using artificial intelligence.
Part 1. Classification of partially edentulous arches using a convolutional
neural network
Toshihito Takahashi a*, Kazunori Nozaki b, Tomoya Gonda a , Kazunori Ikebe a
a
Department of Prosthodontics, Gerodontology and Oral Rehabilitation, Osaka University Graduate School of Dentistry.Japan
b
Division of Medical Information, Osaka University Dental Hospital.Japan

Abstract
Purpose: The purpose of this study was to develop a method for classifying dental arches using a convolutional neural network (CNN) as the first step in a
system for designing removable partial dentures.
Methods: Using 1184 images of dental arches (maxilla: 748 images; mandible: 436 images), arches were classified into four arch types: edentulous, intact
dentition, arches with posterior tooth loss, and arches with bounded edentulous space. A CNN method to classify images was developed using Tensorflow and
Keras deep learning libraries. After completion of the learning procedure, the diagnostic accuracy, precision, recall, F-measure and area under the curve (AUC)
for each jaw were calculated for diagnostic performance of learning. The classification was also predicted using other images, and percentages of correct
predictions (PCPs) were calculated. The PCPs were compared with the Kruskal-Wallis test (p = 0.05).
Results: The diagnostic accuracy was 99.5% for the maxilla and 99.7% for the mandible. The precision, recall, and F-measure for both jaws were 0.25, 1.0 and
0.4, respectively. The AUC was 0.99 for the maxilla and 0.98 for the mandible. The PCPs of the classifications were more than 95% for all types of dental arch.
There were no significant differences among the four types of dental arches in the mandible.
Conclusions: The results of this study suggest that dental arches can be classified and predicted using a CNN. Future development of systems for designing
removable partial dentures will be made possible using this and other AI technologies.
Keywords: Removable partial denture, Machine learning, Artificial intelligence, Convolutional neural network

Received date: 2 July 2019 , Accepted date: 28 April 2020, J-STAGE Advance published date: 9 September 2020

1. Introduction Therefore, there is a need for an automated system for designing RPDs
based on the oral conditions of the patient, and that is not dependent on the
The design of removable partial dentures (RPDs) influences not clinician’s knowledge and experience.
only the outcome of the prosthesis itself, but also the prognosis of In recent years, artificial intelligence (AI) technologies have been
the abutment and residual teeth [1, 2]. The design also influences applied in various fields, obviating the need for human input in some
oral functions, such as mastication, pronunciation, and swallowing, cases. In the medical field, studies and clinical applications using AI have
and general quality of life factors, such as enjoyment of food [3] and been developed in the diagnosis of breast cancer [5] and leukemia [6],
cognitive function [4]. For these reasons, clinicians bear a heavy yielding results comparable to or superior to diagnoses made by humans.
responsibility to design a high quality RPD for each patient, using In the dental field, AI has been applied in the diagnosis of dental caries
all the available information about RPD design. However, clinicians [7] and oral cancer [8] with accuracy comparable to human beings.
are usually limited by their inadequate knowledge and experience of These applications have involved the use of machine learning, in which
designing RPDs, and many clinicians will design different RPDs even a convolutional neural network (CNN) is developed to make predictions
if the oral situation is the same. Hence, quality of RPDs is affected from various target images.
by the clinicians’ knowledge, experience or skills of RPD treatment. Several trials have been conducted to develop an AI system for
designing RPDs. When designing RPDs using AI, the computer itself
* Corresponding author at: Department of Prosthodontics, Gerodontology and Oral
needs to acquire information about the dental arch and the residual teeth to
Rehabilitation, Osaka University Graduate School of Dentistry, 1-8 Yamadaoka, Suita, formulate predictions.
Osaka 565-0871, Japan. The aim of this study was to develop a system for designing RPDs using
E-mail address: [email protected] (T.Takahashi). AI involving the classification of the dental arch using a CNN as the first
https://doi.org/10.2186/jpr.JPOR_2019_354 step.
1883-1958/© 2020 Japan Prosthodontic Society. All rights reserved.
116 T. Takahashi et al. / journal of prosthodontic research 65 ( 2021 ) 115–118

2. Materials and Methods Table 1. Diagnostic accuracy, precision, recall, F-measure, and AUC of the maxilla
and mandibula.
2.1. Data collection diagnostic accurary(%) precision recall F-measure AUC
maxilla 99.5 0.25 1 0.4 0.99
A total of 1184 oral photographic images (748 maxillary and 438 mandibula 99.7 0.25 1 0.4 0.98
mandibular) were obtained from patients. The images consisted of four
types of dental arches: edentulous, arches with posterior tooth loss (distal
extension missing), arches with bounded edentulous space (intermediate design of the RPD were decided by clinicians.
missing), and intact dentition (without missing) in each jaw. These images Although this new system is also programmed with information
were judged the arch types previously by authors using the same oral about past RPDs provided by clinicians, the computer itself creates the
images. They were JPEG files which were resized to 224 × 224 pixels. rules for the design of RPDs via autonomous learning. This system can
The images were randomly divided into two datasets: one for training archive information through AI technology, such as big data and machine
(1016 images; 656 maxillary and 360 mandibular) and one for testing learning, and predict the design of RPDs from the results. The flow of this
(168 images; 92 maxillary and 76 mandibular). This study protocol was classification system is as below. The computer reads the training dataset
approved by the Ethical Committee (H30-E26). which is already classified into true arch types and creates (predicts) the
classification rule by learning them. The testing dataset is used to test the
2.2. Algorithm of the convolutional neural network predicted rule and the accuracy of learning is improved by learning again
and again. After finishing the learning, the computer classifies new images
In a deep CNN procedure, Keras was used as the library with of dental arch with prediction rate based on the rule. Clinicians finally have
Tensorflow as the backend. The ImageNet pretrained model of a 152-layer to judge whether the prediction is correct or not.
residual network model (ResNet152) [9] with fine tuning was used for Basic information about the intraoral condition, such as the location
preprocessing, and the dataset was trained to classify the type of dental of missing teeth, and the prosthetic situation of the residual teeth and
arch. The training dataset was separated into 10 batches for every epoch occlusion, is necessary for the computer to learn and create the design of
and 50 epochs were run at a learning rate of 0.01. an RPD. Therefore, in this study, the dental arches were classified using
oral photographic images and a CNN (an AI technology) as the first step in
2.3. Statistical analysis developing a system for designing RPDs, and the diagnostic accuracy of
learning and the prediction of dental arches was evaluated.
The diagnostic accuracy, precision, recall, F-measure, and the area When evaluating the effectiveness of machine learning, diagnostic
under the receiver operating characteristic (ROC) curve (i.e., the AUC) of accuracy was the most important factor, and the ideal value for diagnostic
each jaw were calculated to assess the test dataset. After completion for accuracy was 100%. The ideal values of other indicators, such as recall,
the learning procedure, predictions of the dental arch in both jaws were precision, and, importantly, the F-measure, were 1.0. The AUC is an index
conducted using 20 other images of each arch type, and the percentages of of learning performance, and the ideal value was close to 1.0. In this study,
correct prediction (PCPs) were recorded. Comparisons were made of the the diagnostic accuracy of learning was more than 99% and the AUC was
median of PCPs for each arch type using the Kruskal-Wallis test with a more than 0.98 in both jaws. These accuracy values were higher than those
post-hoc comparison using the Bonferroni method. All statistical analyses of previous studies using a CNN [12], and a value of more than 90% was
were performed using SPSS statistics Ver22 (IBM, Armonk, NY). considered suitable for application in a clinical setting. Although recall
values for both jaws were almost 1.0, the precision and F-measure values
3. Results were limited to 0.25 and 0.4, respectively. These results indicate that
correct answers were fully judged as correct, but several wrong answers
In the maxilla, diagnostic accuracy was 99.5%, precision was 0.25, were also judged as correct in this system. In other words, this system still
recall was 1.0, F-measure was 0.4, and AUC was 0.99. In the mandible, contains some uncertainties, despite the high diagnostic accuracy and AUC.
diagnostic accuracy was 99.7%, precision was 0.25, recall was 1.0, To improve the quality of the learning for use in clinical settings, wrong
F-measure was 0.4, and AUC was 0.98 (Table 1). answers need to be correctly judged as wrong and therefore the value of
Figure 1 shows an example of PCP results in the maxilla. In these the F-measure must be raised to at least 0.8.
images, the PCPs of edentulous, distal extension missing, and without The learning procedure of this system consisted of 50 epochs, a
missing group were 100%, and that of intermediate missing group was batch size of 10 and a learning rate of 0.01. The number of epochs is an
99.8%. The median of PCP results in the maxilla were 100.0% (99.7-100.0) important factor in the accuracy of deep learning and, in general, the higher
(edentulous), 99.4% (99.2-99.9) (distal extension missing), 97.5% (95.1- the number of epochs, the more accurate the learning will be. The batch
99.3) (intermediate missing), and 97.9% (97.7-99.9) (without missing). size and learning rate are also important factors in the diagnostic accuracy.
There was a significant difference between edentulous and intermediate The greater the batch size, the longer the learning time will be. In the
missing group in the maxilla (p = 0.017) (Fig. 2). In the mandible, the pre-learning procedure to decide learning conditions, several trials were
median of PCP results were 100.0% (99.7-100.0) (edentulous), 99.3% conducted with different epoch numbers and batch sizes. As the result,
(96.3-100.0) (distal extension missing), 98.8% (97.9-99.9) (intermediate more than 50 epochs did not increase the accuracy, and a batch size of
missing), and 98.8% (96.4-100.0) (without missing). There were no more than 10 only prolonged the learning time. A learning rate of 0.01 was
significant differences among the four types of dental arches in the usually used in past studies using CNN [12]. Other rates were also tried
mandible (p = 0.358) (Fig. 2). in pre-learning procedure and resulted in lower accuracy. Therefore, these
three values were selected in this study.
4. Discussion In this classification system, each image was presumably classified
into arch type according to color distribution. In the edentulous group,
The development of a system for designing RPDs using AI technology an image is consisted of only red color of residual ridge, palate and
was firstly reported in 1980s as an “expert system” [10, 11]. Such systems floor of mouth. On the other hand, in other three groups it is consisted
were programmed by clinicians with special knowledge, techniques, and of combination of white color of residual teeth, metallic colors of
enough experience with RPDs so that, in a real sense, the rules for the restorations and red color, and it is classified into each dental arch
 T. Takahashi et al. / journal of prosthodontic research 65 ( 2021 ) 115–118 117

Fig. 1. Photographic images and graphs showing an example of the percentage of correct predictions (PCPs) in the maxilla. a: edentulous, b: arches with posterior tooth
loss (distal extension missing), c: arches with bounded edentulous space (intermediate missing), and d: intact dentition (without missing).

Fig. 2. Graphs of the median of percentages of correct predictions (PCPs). * denotes a significant difference (p < 0.05). a: maxilla, b: mandible. Edentulous: edentulous,
distal extension: arches with posterior tooth loss, intermediate: arches with bounded edentulous space, and without missing: intact dentition.

according to this color distribution. The results of prediction were higher were still significant differences between the intermediate missing teeth
in edentulous and without missing because color distribution was simple and the other arch types in the maxilla. To improve the accuracy, more
and uniform, but those of other two arches were lower because color images are required, especially of intermediate missing in the maxilla.
distribution was various depend on the distribution of Our finding that the PCPs were greater than 95% in all types of dental
missing teeth and residual teeth. Especially in the intermediate missing arches suggests that this system can be applied and a system for designing
group, there were more patterns of missing teeth and residual teeth than in RPDs can be developed when combined with other systems using
the other three arch types. This variation made it difficult to make accurate artificial intelligence. Further steps to be developed in the future include
prediction and the difference of prediction accuracy in intermediate the development of image recognition using photographic images of
missing group between two jaws. The number of images of maxilla was components of RPDs and panoramic X-rays or images of residual teeth.
much more than that of mandibula and therefore, the patterns of missing
teeth and residual teeth in the intermediate missing group for maxilla were 5. Conclusion
more various than those of mandibula. This made the prediction accuracy
of maxilla lower, and difference between two jaws was occurred. Within the limitations of this study, the following conclusions were
In this study, images used in both learning were selected without drawn.
special exclusion criteria in order to correct images as many as possible. 1. The diagnostic accuracy was 99.5% for the maxilla and 99.7% for the
Therefore, there were some images with problems such as small lacking mandible.
of dental arch, out of focus and these images may reduce the learning 2. The percentage of correct predictions was more than 95% in all
performance. In the case of prediction procedure, when using imperfect dental arches. In the maxilla, there was a significant difference between
images aforementioned above, computer will recognize them as they are edentulous and arches with bounded edentulous space. No significant
and predict the dental arch. As the result,omputer will predict them as differences were observed among four arch types in the mandible.
wrong type of dental arch. Considering these, complete images in these
points should be necessary to predict correctly. Conflicts of interest
The PCPs of this study were more than 95% in all dental arches.These
scores are almost comparable with the visual ability of human beings. The authors report no conflicts of interest related to this study.
Although the PCPs were much higher than those of our preliminary
studybecause of the use of new architecture with greater accuracy, there
118 T. Takahashi et al. / journal of prosthodontic research 65 ( 2021 ) 115–118

References [7] L  ee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries
using a deep learning-based convolutional neural network algorithm. J Dent
2018; 77: 106-11.
[1] T
 ada S, Allen PF, Ikebe K, Zheng H, Shintani A, Maeda Y. The impact of the [8] J eyaraj PR, Samuel Nadar ER. Computer-assisted medical image classification
crown-root ratio on survival of abutment teeth for dentures. J Dent Res 2015; for early diagnosis of oral cancer employing deep learning algorithm. J Cancer
94: 220S-5S. Res Clin Oncol 2019; 145: 829-37.
[2] M izuno Y, Gonda T, Takahashi T, Tomita A, Maeda Y. Root Fracture of [9] He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition.
Abutment Teeth for Partial Removable Dental Prostheses. Int J Prosthodont arXiv 2015: 1512. 03385.
2016; 29: 461-6. [10] M aeda Y, Tsutsumi S, Okada M, Minoura M, Nobuki T, Okuno Y. An
[3] Inomata C, Ikebe K, Okubo H, Takeshita H, Mihara Y, Hatta K, et al. Dietary expert system for designing removable partial dentures (Part 1) Selection of
intake is associated with occlusal force rather than number of teeth in 80-y-old Abutments. J Jpn Prosthodont Soc 1987; 29: 833-8.
Japanese. JDR Clin Trans Res 2017; 2: 187-97. [11 ] Maeda Y, Tsutumi S, Minoura M, Okada M, Nokubi T, Okuno Y. An expert
[4] Hatta K, Ikebe K, Gondo Y, Kamide K, Masui Y, Inagaki H, et al. Influence system for designing removable partial dentures. Part 2: the role of the data base
of lack of posterior occlusal support on cognitive decline among 80-year-old constructed with clinical cases. J Jpn Prosthodont Soc 1985; 31: 1531-35.
Japanese people in a 3-year prospective study. Geriatr Gerontol Int 2018; 18: [12] Anwar SM, Majid M, Qayyum A, Awais M, Alnowami M, Khan MK. Medical
1439-46. Image Analysis using Convolutional Neural Networks: A Review. J Med Syst
[5] Zhu Z, Albadawy E, Saha A, Zhang J, Harowicz MR, Mazurowski MA. Deep 2018; 42: 226.
learning for identifying radiogenomic associations in breast cancer. Comput
Biol Med 2019; 109: 85-90.
[6] R ehman A, Abbas N, Saba T, Rahman SIU, Mehmood Z, Kolivand H.
Classification of acute lymphoblastic leukemia using deep learning. Microsc
Res Tech 2018; 81: 1310-7.