Deep Learning Based on Swin-Transformer and 3D U-Net: Implant Three-Dimensional Position Planning

Area of Science:

• * Artificial Intelligence in Dentistry
• * Medical Imaging Analysis
• * Oral and Maxillofacial Surgery

Background:

• * Accurate identification of anatomical landmarks in the posterior mandible is crucial for safe dental implant placement.
• * Mandibular lingual concavities and the mandibular nerve canal pose significant risks if not properly identified.
• * Current methods for implant planning can be time-consuming and may lack precision.

Purpose of the Study:

• * To develop a deep learning model for automated identification of mandibular lingual concavities.
• * To predict precise, biologically guided three-dimensional (3D) implant positions.
• * To assess the model's performance in segmenting anatomical structures and classifying concavities.

Main Methods:

• * A deep learning framework using 3D U-Net and Swin-Transformer was developed.
• * The model processed cone-beam computed tomography (CBCT) images of patients with posterior mandibular edentulism.
• * Automated segmentation of teeth, mandible, and mandibular nerve canal, plus classification of lingual concavities and implant key point prediction were performed.

Main Results:

• * High accuracy was achieved in dental segmentation (Dice coefficients 0.87-0.91) and lingual concavity classification (0.92-0.97).
• * Predicted implant positions maintained a safety margin of 3.20-3.89 mm from the mandibular nerve canal.
• * Sufficient bone volume was preserved, with average buccal/lingual cervical bone widths of 4.94-5.78 mm.

Conclusions:

• * The deep learning model demonstrated robust performance in identifying anatomical structures and predicting implant positions.
• * The model offers a clinically acceptable safety margin for implant placement in internal validation.
• * This AI framework aids in mitigating intraoperative complications and reducing risks of postoperative complications.