Panoramic Insights: Predicting Age and Gender from Dental X-Rays Using Deep Learning Models.

Area of Science:

• Forensic Odontology
• Artificial Intelligence
• Radiology

Background:

• Manual age and gender estimation from dental radiographs is labor-intensive and subjective.
• Deep learning (DL) offers automated, efficient, and objective assessment capabilities.
• Convolutional Neural Network (CNN) architectures are well-suited for image analysis tasks.

Purpose of the Study:

• To predict age and gender using various deep learning models.
• To compare the efficiency of different CNN architectures for demographic estimation.
• To evaluate the performance of individual and ensemble DL models.

Main Methods:

• Utilized 2341 panoramic radiographs for training and testing DL models.
• Employed preprocessing techniques: normalization, resizing (224x224), and data augmentation (AutoAugment, RandAugment).
• Implemented transfer learning with pretrained ResNet and DenseNet models, incorporating an ensemble strategy.

Main Results:

• DenseNet161 achieved the highest individual accuracy: 90.4% for gender and 94% for age.
• Ensemble models reached 94% accuracy for gender and 90.4% for age.
• Ensemble learning enhanced gender prediction but did not surpass DenseNet161 for age estimation.

Conclusions:

• Deep learning models provide a promising, objective method for demographic estimation from panoramic radiographs.
• DenseNet161 demonstrates high efficacy for age and gender classification.
• Further validation with larger, multi-center datasets and standardized protocols is necessary for clinical application.