Sex estimation from maxillofacial radiographs using a deep learning approach.

Area of Science:

• Radiology
• Artificial Intelligence
• Forensic Anthropology

Background:

• Accurate sex estimation is crucial in forensic and anthropological analyses.
• Traditional methods can be time-consuming and require specialized expertise.
• Deep learning offers potential for automated and efficient analysis of medical images.

Purpose of the Study:

• To develop and evaluate deep learning models for sex estimation using lateral cephalograms.
• To compare the performance of VGG16 and DenseNet-121 models for this task.
• To investigate the interpretability of model predictions using saliency maps.

Main Methods:

• Retrospective analysis of 600 lateral cephalograms.
• Implementation of two deep learning architectures: VGG16 and DenseNet-121.
• Generation of saliency maps using gradient-weighted class activation mapping (Grad-CAM).

Main Results:

• Both VGG16 and DenseNet-121 achieved high performance across multiple metrics (accuracy, sensitivity, precision, F1 score, AUC).
• Saliency maps demonstrated distinct spatial patterns differentiating male and female images.
• Model-specific differences in saliency map locations were observed, independent of sex.

Conclusions:

• Deep learning models can accurately and reliably estimate sex from lateral cephalograms.
• Saliency map analysis provides insights into the features utilized by the models for sex determination.
• This approach holds promise for improving efficiency and objectivity in sex estimation.