A potential method for sex estimation of human skeletons using deep learning and three-dimensional surface scanning

Area of Science:

• Forensic Anthropology
• Computer Vision
• Medical Imaging

Background:

• Deep learning shows promise in forensic anthropology, but radiological methods are costly and immobile.
• Portable systems are needed for efficient and accurate skeletal sex estimation.

Purpose of the Study:

• To develop and evaluate a portable system for pelvic sex estimation using deep learning and 3D surface scanning.
• To compare the accuracy of the developed system against human experts.

Main Methods:

• Convolutional neural networks (CNNs) were trained on CT-derived pelvic images (ventral pubis, dorsal pubis, greater sciatic notch).
• Trained CNN models were applied to images from a handheld 3D surface scanner.
• A single-blind trial compared the system's accuracy to that of two anthropologists.

Main Results:

• CNN models achieved high accuracy (98.0-98.5%) for ventral and dorsal pubis estimation from CT data.
• Models showed over 95% accuracy for ventral and dorsal pubis estimation using 3D scanning data.
• The combined CNN model achieved 97.1% accuracy with 3D scanning, surpassing human experts.

Conclusions:

• Deep learning integrated with 3D surface scanning offers a rapid and accurate method for skeletal sex estimation.
• This portable system has significant potential for forensic anthropology applications.
• The technology demonstrates superior performance compared to traditional methods and human analysis.