DeepTooth: Estimating Age and Gender with Panoramic Radiograph Image
Issued Date
2023-01-01
Resource Type
Scopus ID
2-s2.0-85169797533
Journal Title
2023 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2023
Rights Holder(s)
SCOPUS
Bibliographic Citation
2023 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2023 (2023)
Suggested Citation
Somdej W., Thamvongsa A., Hirunchavarod N., Sributsayakarn N., Pornprasertsuk-Damrongsri S., Jirarattanasopha V., Intharah T. DeepTooth: Estimating Age and Gender with Panoramic Radiograph Image. 2023 International Technical Conference on Circuits/Systems, Computers, and Communications, ITC-CSCC 2023 (2023). doi:10.1109/ITC-CSCC58803.2023.10212499 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/89994
Title
DeepTooth: Estimating Age and Gender with Panoramic Radiograph Image
Author's Affiliation
Other Contributor(s)
Abstract
Age estimation is one of forensic science's most important steps for personal identification. As a durable tissue, dental characteristics assessed from radiographs have been used to estimate the chronological age. However, current age estimation methods from dental radiographs are complicated, time-consuming, and highly dependent on manual estimation, which is prone to error. In this research, we developed models for estimating the age and gender of humans from radiographic images using the EfficientNet called DeepTooth model. This study proposes one classification model for gender classification, one regression model for age estimation, and three classification models for age estimation (one model trained from both genders and the other two trained from only males or females). For age estimation, the classification and regression models trained from both genders achieved RMSE values of 5.09 and 2.26, respectively, while the model trained from male or female achieved an average of 4.74. For gender classification, we used the same backbone and data-splitting strategy. The model accuracy was 70.32 percent.