Related Experiment Video
Updated: Jun 5, 2025
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
2.6K
Deep learning versus human assessors: forensic sex estimation from three-dimensional computed tomography scans
Ridhwan Lye1, Hang Min2,3, Jason Dowling4,5
1Centre for Forensic Anthropology, School of Social Sciences, The University of Western Australia, Perth, Australia. ridhwan.dawudlye@research.uwa.edu.au.
Scientific Reports
|December 3, 2024
Summary
A new deep learning (DL) framework significantly improves cranial sex estimation accuracy. This AI approach, tested on Indonesian CT scans, reduces human bias and outperforms traditional methods for forensic anthropology.
Area of Science:
- Forensic Anthropology
- Computer Science
- Medical Imaging
Background:
- Traditional cranial sex estimation relies on subjective visual assessments by forensic anthropologists.
- Existing methods can exhibit human bias and reduced accuracy across diverse populations.
Purpose of the Study:
- To develop and evaluate an automatic deep learning (DL) framework for enhanced cranial sex estimation.
- To reduce bias and improve accuracy compared to human observers in forensic anthropology.
Main Methods:
- Utilized 200 cranial CT scans from Indonesian individuals.
- Trained and evaluated various DL network configurations.
- Compared DL performance against a human observer using established standards.
Main Results:
- The most accurate DL network achieved 97% classification accuracy, surpassing the human observer's 82%.
- The DL model learned sex estimation alongside cranial traits as an auxiliary task.
- Grad-CAM visualisations showed the DL model focused on specific cranial traits, size, and shape.
Conclusions:
- Deep learning offers a promising tool to assist forensic anthropologists in sex estimation.
- DL frameworks can provide more accurate and less biased skeletal sex estimations.
- This technology has the potential to improve the reliability of forensic analyses.
