MorphMaskFormer: a transformer-based deep segmentation model for multi-class Demirjian stage estimation from...

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Forensic Odontology

Background:

• Dental age estimation is crucial for clinical and forensic applications.
• Current methods for assessing third-molar development can be subjective and time-consuming.
• Automated tools are needed to improve accuracy and efficiency.

Purpose of the Study:

• To develop an advanced deep learning model for automated determination of third-molar developmental stages.
• To utilize the Demirjian classification for precise dental age estimation.
• To enhance the accuracy and objectivity of age assessment in panoramic radiographs.

Main Methods:

• A deep learning model, MorphMaskFormer, was developed based on the UNet architecture with a transformer attention module.
• The model performed binary and multi-class segmentation of third-molar developmental stages (Demirjian A-H) on 888 panoramic radiographs.
• Performance was evaluated using IoU, Dice coefficient, Precision, Recall, and inference time, compared to baseline models.

Main Results:

• MorphMaskFormer achieved a Dice score of 0.9461 and IoU of 0.8985, outperforming all baseline models.
• The model demonstrated the fastest inference time at 78.59 ms.
• High accuracy was observed for stages A, D, and H, with an overall component accuracy of 72.41%.

Conclusions:

• MorphMaskFormer enables precise pixel-level segmentation of dental developmental stages.
• The model reduces inter-observer variability and evaluation time.
• Its accuracy and efficiency make it a valuable tool for clinical and forensic age estimation.