Related Experiment Video
Updated: Aug 25, 2025
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.0K
Sex determination from lateral cephalometric radiographs using an automated deep learning convolutional neural
Maryam Khazaei1, Vahid Mollabashi2, Hassan Khotanlou3
1Department of Biostatistics, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Iran.
Imaging Science in Dentistry
|October 14, 2022
Summary
Automated sex determination using convolutional neural networks (CNNs) achieved 90% accuracy on lateral cephalometric radiographs. This AI-driven approach minimizes human error for reliable sex identification.
Area of Science:
- Radiology
- Artificial Intelligence
- Forensic Anthropology
Background:
- Traditional morphometric and anthropometric methods for sex identification are prone to observer error.
- Automated methods offer a potential solution to improve accuracy and reduce subjectivity.
Purpose of the Study:
- To investigate the efficacy of convolutional neural networks (CNNs) for automated sex determination.
- To utilize lateral cephalometric radiographs as input for CNN-based sex identification.
Main Methods:
- A dataset of 1,476 lateral cephalometric radiographs from Iranian subjects (18-49 years) was used.
- Three CNN architectures (DenseNet, ResNet, VGG) were trained and tested, with 80% of data for training and 20% for testing.
- Hyperparameter tuning, preprocessing, and data augmentation were performed; transfer learning was employed.
Main Results:
- The DenseNet121 architecture achieved the highest accuracy (90%) in sex determination.
- Prediction accuracy was comparable for both male and female subjects.
- Transfer learning consistently enhanced predictive performance across all evaluated architectures.
Conclusions:
- CNNs can accurately determine sex from lateral cephalometric radiographs, independent of human bias.
- Automated feature extraction by CNNs reduces subjectivity in sex identification.
- Further research with larger datasets is recommended for broader application.
