Automated cephalometric landmark detection with confidence regions using Bayesian convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Orthodontics

Background:

• Cephalometric analysis is crucial in orthodontics but faces challenges with landmark tracing reliability and accuracy.
• Existing automatic plotting systems often lack the reliability needed for clinical applications, particularly for specific landmarks.

Purpose of the Study:

• To develop a novel framework for locating cephalometric landmarks with confidence regions.
• To enhance the accuracy and reliability of landmark detection using Bayesian Convolutional Neural Networks (BCNN).

Main Methods:

• A Bayesian Convolutional Neural Network (BCNN) framework was developed for cephalometric landmark detection.
• The model was trained on dental X-ray images from the ISBI 2015 challenge, incorporating region of interest extraction and uncertainty estimation.
• Post-processing methods were applied to pixel probabilities and uncertainties from the Bayesian model predictions.

Main Results:

• The framework achieved a mean landmark error (LE) of 1.53 ± 1.74 mm.
• Successful detection rates (SDR) were 82.11%, 92.28%, and 95.95% within 2, 3, and 4 mm ranges, respectively.
• Significant error reduction was observed for the Gonion landmark, and improved performance in identifying anatomical abnormalities was noted.

Conclusions:

• The developed framework provides cephalometric landmarks with confidence regions, addressing previous accuracy limitations.
• This tool can serve as a valuable computer-aided diagnosis aid and educational resource in orthodontics.
• The confidence regions enhance clinical convenience and support better treatment decisions.