Fig 1.
Flowchart of study participants and design.
A total of 520 Japanese children were recruited. They were assessed for eligibility. Finally, 520 children were registered as participants. During this study, 1 participant was excluded (not meeting the inclusion criteria). After data cleansing, three datasets were created to build each model to determine the presence of deep bite, maxillary protrusion, and crowding, respectively.
Table 1.
Each variable for the dataset.
Fig 2.
Guide frame that appears on the smartphone application.
The guardians captured three images (front, left side, right side) of the child’s tooth alignment according to the guide frame on the smartphone application.
Fig 3.
Representative photographic image.
Representative photographic images of the frontal view (a), left side (b), and right side (c) of the child’s tooth alignment.
Table 2.
Breakdown of participants by residential area.
Table 3.
Breakdown of participants by annual household income.
Fig 4.
Explicability results of the deep bite classification model.
(a) PI. (b) Representative photographic images of activation maps (frontal image). Red areas indicate areas of particular focus for the AI model to make decisions. The model was built using the Elastic-Net Classifier (L2/ Binomial Deviance) algorithm.
Fig 5.
Explicability results of the maxillary protrusion classification model.
(a) PI. (b) Representative photographic images of activation maps (frontal and left side images). Red areas indicate areas of particular focus for the AI model to make decisions. The model was built using the Elastic-Net Classifier (mixing alpha = 0.5/ Binomial Deviance) algorithm.
Fig 6.
Explicability results of the crowding classification model.
(a) PI. (b) Representative photographic images of activation maps (frontal image). Red areas indicate areas of particular focus for the AI model to make decisions. The Nystroem Kernel SVM Classifier algorithm was employed to build the model.
Table 4.
Performance metrics of the machine learning models on the test dataset.
Table 5.
Performance metrics of the logistic regression models on the test dataset.
Table 6.
Association between deep bite and oral habits.
Table 7.
Association between maxillary protrusion and oral habits.
Table 8.
Association between crowding and oral habits.