Abstract
Dental age estimation is a politically and ethically sensitive area of forensic odontology with great implications for individuals who are incorrectly classified: potential violation of a minor’s rights, being denied access to appropriate services or adults in the system being treated as children such as in asylum seeker cases. The aims of this project are: To investigate the use of artificial intelligence (AI) on dental panoramic tomography radiographs to predict the age of an individual to be under or over 16; to evaluate how accurate AI is at correctly classifying images in the context of dental age estimation; to compare the performance of AI in estimating the dental of age males and females.An observational analytics cross-sectional study was performed with a sample of 5040 radiographs of Brazilian subjects with 4200 used in training (n = 4200; 2100 males and 2100 females) between the age of 6 and 22.9 years, 16 years being in the middle of this range. The images were used to train and validate AI software (DenseNet121) to recognise patterns of the lower left mandibular third molar to classify the radiographs into two categories: over 16 or under 16 years old.
A stochastic optimisation algorithm (SGD) was used for optimisation during the training of the network. The images dataset was divided into 5 equal subsets, with four being used for training and the fifth to estimate the parameters and therefore compute the accuracy of the model. Results showed that DenseNet121 software could accurately estimate the age of both males (88% accuracy) and females (83% accuracy) by assigning them a binary result of either over 16 or under 16 years old. Confusion matrices and AUC graphs showed that classification by AI was more sensitive, specific and precise for males compared with females.
AI for age estimation using lower third molars was accurate with minimal difference between males and females, suggesting great potential of future development and use of AI in this field.
| Date of Award | 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Scheila Manica (Supervisor), Hemlata Pandey (Supervisor), Ademir Franco do Rosario Junior (Supervisor) & Alex Gardner (Supervisor) |