Automatic 3-Dimensional Cephalometric Landmarking via Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Orthodontics and Maxillofacial Surgery

Background:

• 3D imaging is increasingly vital for assessing complex dentofacial deformities and planning orthognathic surgeries.
• Accurate 3D cephalometric analysis is critical but current automatic landmark localization methods lack robustness and generalizability.
• Manual landmarking by highly trained operators remains the standard, posing time and resource challenges.

Purpose of the Study:

• To train and evaluate a deep learning (DL) pipeline using SpatialConfiguration-Net for automatic 3D cephalometric landmark localization on computed tomography (CT) scans.
• To assess the accuracy and reliability of the DL model compared to manual landmarking.

Main Methods:

• A retrospective diagnostic study utilizing a dataset of 198 presurgical CT scans (160 training/validation, 38 test).
• Manual localization of 33 landmarks by operators served as reference data.
• The DL pipeline's performance was evaluated based on localization accuracy, cephalometric measurements, and comparison with manual landmarking reproducibility.

Main Results:

• The DL model achieved a mean localization error of 1.0 ± 1.3 mm on the test set, with success detection rates of 90.4% (2.0 mm), 93.6% (2.5 mm), and 95.4% (3.0 mm).
• Mean errors for angular and linear cephalometric measurements were -0.3 ± 1.3° and -0.1 ± 0.7 mm, respectively.
• Measurements derived from DL predictions were within Bland-Altman 95% limits of agreement for 91.9% of skeletal and 71.8% of dentoalveolar variables compared to manual reproducibility.

Conclusions:

• The developed DL method demonstrates high accuracy in 3D cephalometric landmark localization on CT scans, despite requiring further improvement.
• The model's reliability for skeletal evaluation is comparable to that achieved by clinicians.
• This DL pipeline shows significant potential to enhance efficiency and accuracy in 3D cephalometric analysis for clinical applications.