Deep learning for sex determination: Analyzing over 200,000 panoramic radiographs

Area of Science:

• Dentistry
• Artificial Intelligence
• Radiology

Background:

• Sex determination from dental records is crucial for identification.
• Panoramic radiographs (PR) offer a comprehensive view of dental structures.
• AI, particularly convolutional neural networks (CNNs), shows promise in analyzing medical images.

Purpose of the Study:

• To evaluate an AI tool's effectiveness for sex determination using PR.
• To identify factors influencing the performance of AI models in sex classification.
• To assess the impact of image resolution, age, and dentition on AI accuracy.

Main Methods:

• Utilized a dataset of 207,946 PRs from 15 clinical centers in Brazil.
• Preprocessed data included anonymization and extraction of demographic and dental information.
• Trained and validated two CNN architectures (standard CNN and ResNet) using hyperparameter tuning and cross-validation.

Main Results:

• The standard CNN model achieved 95.02% accuracy in sex estimation.
• Image resolution was identified as a key factor influencing AI performance.
• The ResNet model demonstrated high accuracy (>96%) in individuals over 16 years, with better performance on female images.

Conclusions:

• AI-powered sex determination from PR is effective and accurate.
• Image resolution, patient age, and sex are significant factors affecting AI performance.
• This AI tool has potential applications in forensic odontology and patient identification.