Development of a Deep Learning Model to Automatically Identify Palatal Landmarks on Three-Dimensional Maxillary...

Area of Science:

• Artificial Intelligence in Dentistry
• 3D Imaging and Analysis
• Craniofacial Morphology

Background:

• Accurate identification of palatal landmarks on 3D digital maxillary dental casts is crucial for morphological assessment.
• Manual annotation of these landmarks is time-consuming and prone to variability.
• Developing automated methods can improve efficiency and consistency in dental research and clinical practice.

Purpose of the Study:

• To develop a deep learning model for automatic identification of 12 palatal landmarks on 3D digital maxillary dental casts.
• To evaluate the performance and accuracy of the developed deep learning model.

Main Methods:

• A dataset of 377 3D digital maxillary dental casts was used, with 12 palatal landmarks manually annotated as ground truth.
• A two-stage PointNet++ deep learning architecture was employed for landmark detection.
• Model accuracy was assessed by measuring the linear discrepancy between predicted and ground-truth landmark positions.

Main Results:

• The developed deep learning model achieved a mean landmark detection error of 0.55 mm (±0.49 mm).
• The model demonstrated high predictive accuracy, identifying 90% of landmarks within 1 mm and 98% within 2 mm of the ground truth.
• The hierarchical model effectively extracted both local and global features from the point cloud data.

Conclusions:

• A high-performing deep learning model for automated palatal landmark identification on 3D dental casts has been successfully developed.
• This automated approach significantly reduces manual annotation time, enhancing efficiency in clinical and research settings.
• The model facilitates more efficient morphological assessment of the palate.