Automated 3D cephalometry: A lightweight V-net for landmark localization on CBCT

Area of Science:

• Medical Imaging and Artificial Intelligence
• Orthodontics and Dental Diagnostics

Background:

• Cephalometric analysis is crucial for orthodontic clinical decision support, traditionally requiring manual landmark identification on 3D cone beam CT (CBCT) scans.
• Manual analysis is time-consuming and operator-dependent, highlighting the need for automated solutions in clinical workflows.

Purpose of the Study:

• To develop and validate a lightweight deep learning (DL) model for the automatic localization of 16 key anatomical landmarks on CBCT scans.
• To assess the accuracy and reliability of the DL model's landmark identification and subsequent cephalometric measurements compared to manual methods.

Main Methods:

• A V-net deep learning architecture was trained on 350 manually annotated CBCT scans from diverse imaging systems and patient demographics.
• The model's performance was evaluated based on mean landmark localization error and compared with manually derived linear and angular cephalometric measurements.
• Bland-Altman analysis was employed to assess the agreement between automated and manual measurements.

Main Results:

• The DL model achieved a mean landmark localization error of 1.95 ± 1.06 mm, within the clinically acceptable 2 mm threshold.
• Automated cephalometric measurements showed minimal errors (-0.15 ± 0.95° for angular, 0.20 ± 0.28 mm for linear) with strong agreement to manual values.
• Mean inference time was under 32 seconds per scan, demonstrating computational efficiency.

Conclusions:

• The developed lightweight DL model reliably automates cephalometric landmark identification and measurement, offering a viable alternative to manual procedures.
• The model's accuracy, robustness across heterogeneous datasets, and fast inference times support its potential as a clinical decision support tool in orthodontics.
• Automated cephalometric analysis can enhance efficiency and consistency in orthodontic treatment planning, particularly for critical parameters like the ANB angle.