Three-dimensional maxillary virtual patient creation by convolutional neural network-based segmentation on cone-beam...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in dentistry
• 3D reconstruction

Background:

• Cone-beam computed tomography (CBCT) is crucial for dental diagnostics.
• Accurate segmentation of maxillary structures is essential for virtual patient (MVP) creation.
• Current segmentation methods can be time-consuming and require manual refinement.

Purpose of the Study:

• To evaluate integrated convolutional neural network (CNN) models for automated MVP creation.
• To qualitatively and quantitatively assess the accuracy and efficiency of the segmentation process.
• To determine the clinical utility of automated segmentation in maxillofacial applications.

Main Methods:

• Integration of three validated CNN models for segmenting maxillary complex, sinuses, and dentition.
• Qualitative assessment by experts scoring segmentation refinements (0-10 scale).
• Quantitative analysis of automated segmentation (AS) vs. refined segmentation (RS) performance, including time and inter-observer consistency.

Main Results:

• Automated segmentation (AS) achieved an average time of 1.7 minutes.
• Excellent overlap between AS and RS with a Dice Similarity Coefficient (DSC) of 99.3%.
• High inter-observer consistency for refinements with a 95% Hausdorff Distance (HD) of 0.045 mm.

Conclusions:

• Integrated CNN models provide fast, accurate, and consistent MVP creation.
• The system demonstrates strong inter-observer consistency.
• Automated segmentation is a valuable tool for clinical orthodontics, implantology, and maxillofacial surgery.