Transformer and Attention Enhanced Deep Learning Approach for CBCT-Based Mental Foramen Classification and...

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• The precise identification of the mental foramen is crucial for successful dental surgical procedures.
• Variations in the mental foramen's anatomy can pose challenges in clinical practice.

Purpose of the Study:

• To develop and evaluate deep learning models for automated classification and segmentation of the mental foramen.
• To enhance the accuracy and reliability of mental foramen detection in cone-beam computed tomography (CBCT) images.

Main Methods:

• Utilized a unique, manually annotated CBCT dataset for training and validation.
• Developed an ensemble deep learning model (DAFNet) with Transformer and Attention mechanisms for classification.
• Proposed a UNet-based segmentation model (TA-UNet) incorporating Transformer and Attention mechanisms.

Main Results:

• The DAFNet model achieved 95.20% validation accuracy and 91.25% test accuracy (AUC: 0.9648).
• The TA-UNet model demonstrated a Dice score of 0.8804 and IoU of 0.8386.
• Both models outperformed conventional deep learning architectures in detecting and segmenting the mental foramen.

Conclusions:

• Deep learning models can automatically detect the mental foramen with high accuracy in CBCT images.
• Reliable segmentation of the mental foramen is achievable using the proposed TA-UNet model.
• These findings support the integration of AI in dental imaging for improved surgical planning.