Deep Learning for Sex Estimation from Whole-Foot X-Rays: Benchmarking CNNs for Rapid Forensic Identification

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate sex estimation is vital in forensic identification, especially with fragmented remains or when DNA analysis is impractical.
• Traditional methods using single bone measurements have limitations in accuracy.
• Automated sex estimation using advanced imaging techniques remains an underexplored area.

Purpose of the Study:

• To evaluate the efficacy of deep convolutional neural networks (CNNs) for automated sex estimation.
• To utilize entire foot radiographs as the primary data source for sex estimation.
• To compare the performance of various CNN architectures for this task.

Main Methods:

• Retrospective collection of 471 adult foot radiographs (238 male, 233 female).
• Training six CNN architectures (AlexNet, ResNet-18/50, ShuffleNet, GoogleNet, InceptionV3) via transfer learning with data augmentation.
• Performance evaluation using accuracy, sensitivity, specificity, precision, and F1-score.

Main Results:

• The InceptionV3 CNN model achieved the highest accuracy at 97.1%, with high sensitivity (97.5%) and specificity (96.8%).
• ResNet-50 also demonstrated strong performance with 95.7% accuracy.
• These deep learning models significantly surpassed the accuracy of traditional anthropometric methods (typically 72-89%).

Conclusions:

• Deep learning models applied to whole-foot radiographs provide state-of-the-art accuracy for sex estimation.
• This automated approach offers a rapid, reproducible, and cost-effective solution for forensic identification.
• It is particularly valuable in mass disaster scenarios or clinical emergencies where DNA analysis may be delayed or unavailable.