Forensic sex classification by convolutional neural network approach by VGG16 model: accuracy, precision and...

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Artificial Intelligence

Background:

• Sex estimation is vital for biological profile reconstruction in medico-legal identification.
• Traditional methods for sex estimation can be subjective.
• Convolutional Neural Networks (CNNs) present an objective alternative for sex estimation.

Purpose of the Study:

• To evaluate the VGG16 model's reliability for forensic sex prediction.
• To assess the performance of VGG16 using orthopantomography (OPGs).

Main Methods:

• Utilized 1050 OPGs from a dental department.
• Pre-processed and augmented OPG images using Python.
• Evaluated model performance using precision, sensitivity, F1-score, and accuracy, generating heatmaps.

Main Results:

• The VGG16 model achieved an overall accuracy of 89% for sex classification.
• The model demonstrated a balanced performance between sexes with an F1-score of 0.89.
• Highest accuracy (90%) was observed in the 16-20 age group; heatmaps indicated focus on non-anatomical areas.

Conclusions:

• CNNs are accurate for sex classification in medico-legal identification using OPGs.
• The VGG16 model shows potential but requires further research for performance enhancement.
• Future work should incorporate image extraction techniques to improve focus on relevant anatomical areas.