A Deep Learning Framework for Automated Classification and Archiving of Orthodontic Diagnostic Documents

Area of Science:

• Artificial Intelligence in Dentistry
• Deep Learning for Medical Imaging
• Digital Orthodontics Workflow Automation

Background:

• Manual classification and archiving of orthodontic images are time-consuming and error-prone.
• Inaccurate documentation compromises orthodontic treatment accuracy and patient care.
• Need for automated solutions to enhance efficiency and reduce human error in orthodontic diagnostics.

Purpose of the Study:

• To develop and evaluate an AI-driven deep learning framework for automated classification and archiving of orthodontic diagnostic images.
• To improve workflow efficiency and reduce errors in orthodontic image management.
• To lay the groundwork for automated orthodontic diagnosis and treatment planning systems.

Main Methods:

• Utilized a dataset of 61,842 orthodontic images across 13 categories from three dental clinics.
• Implemented a sequential classification approach with primary (extraoral, intraoral, radiographic) and secondary models, enhanced with attention modules.
• Compared the proposed model against pre-trained models (ResNet50, InceptionV3) and validated externally with 13,729 images.

Main Results:

• The AI framework achieved 99.24% accuracy on an external validation set, comparable to human expert performance.
• Demonstrated significantly faster processing times than manual classification methods.
• Gradient-weighted class activation mapping (Grad-CAM) confirmed the model's focus on clinically relevant image features.

Conclusions:

• The developed deep learning framework effectively automates orthodontic diagnostic image classification and archiving.
• The AI system offers high accuracy, clinical relevance, and substantial improvements in processing speed for real-time applications.
• This automated approach reduces healthcare workload and advances digital orthodontics towards automated diagnosis and treatment planning.