An Artificial Intelligence System for Staging the Spheno-Occipital Synchondrosis

Area of Science:

• Dentistry and Oral Health
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate assessment of spheno-occipital synchondrosis (SOS) fusion stages is crucial for orthodontic treatment planning.
• Manual classification of SOS fusion stages from cone beam computed tomography (CBCT) can be time-consuming and subjective.

Purpose of the Study:

• To develop, test, and validate automated deep learning algorithms for classifying SOS fusion stages using CBCT.
• To enhance the accuracy and efficiency of SOS staging through artificial intelligence.

Main Methods:

• Utilized 723 CBCT scans from orthodontic patients.
• Employed deep learning models including ResNet, EfficientNet, and ConvNeXt.
• Developed a novel attention-based model, ConvNeXt+Conv Attention, and integrated YOLOv11 for automated region detection and segmentation.

Main Results:

• The ConvNeXt+Conv Attention model demonstrated superior performance.
• Achieved 88.94% accuracy with manual cropping and 82.49% accuracy in a fully automated workflow.
• The AI pipeline effectively automated SOS fusion stage classification.

Conclusions:

• A novel AI-based pipeline reliably automates SOS fusion stage classification.
• ConvNeXt+Conv Attention achieved the highest accuracy, improving efficiency and consistency in clinical workflows.
• This technology holds significant potential for AI-driven solutions in orthodontics.