Age Estimation using Panoramic Radiographs by Transfer Learning.

Area of Science:

• Radiology
• Dentistry
• Artificial Intelligence

Background:

• Accurate age estimation is crucial in forensic and clinical dentistry.
• Panoramic radiographs offer a non-invasive method for assessing dental development and age.

Purpose of the Study:

• To evaluate the precision of transfer learning models for age estimation using panoramic dental radiographs.
• To introduce a novel methodology for age determination based on these models.

Main Methods:

• Trained four deep learning models (ResNet, EfficientNet, VggNet, DenseNet) on a dataset of 3000 panoramic radiographs.
• Utilized training (n=2400), validation (n=300), and test (n=300) sets with an equal sex and age distribution.
• Calculated mean absolute error (MAE) and root mean square error (RMSE) to assess model performance.

Main Results:

• The EfficientNet-B5 model achieved the best performance with an MAE of 2.83 and RMSE of 4.59.
• Identified key features for age estimation including dentition, maxillary sinus, mandibular body, and mandibular angle.
• Visualized extracted deep learning features across different age groups.

Conclusions:

• Transfer learning models effectively extract age-related features from panoramic radiographs.
• These models provide a viable and accurate method for age estimation in clinical and forensic contexts.