Automated detection of cephalometric landmarks using deep neural patchworks.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Craniofacial Surgery

Background:

• Cephalometric analysis is crucial for diagnosing and planning treatments in orthodontics and orthognathic surgery.
• Manual landmark identification on 3D CT scans is time-consuming and prone to inter-observer variability.
• Automated methods are needed to improve efficiency and accuracy in cephalometric analysis.

Purpose of the Study:

• To evaluate the accuracy of deep neural patchworks (DNPs) for automated identification of 60 cephalometric landmarks on CT scans.
• To determine the potential of DNPs for routine 3D cephalometric analysis in clinical practice.
• To assess the DNP framework's suitability for diagnostics and treatment planning in orthognathic surgery and orthodontics.

Main Methods:

• A deep learning-based segmentation framework, deep neural patchworks (DNPs), was developed.
• DNPs were trained using spherical segmentations of adjacent tissues for 60 cephalometric landmarks.
• Automated landmark identification was performed on a test dataset of 30 adult skull CT scans, with accuracy compared to manual annotations by two clinicians.

Main Results:

• The DNP framework successfully identified all 60 cephalometric landmarks.
• The mean error for DNP landmark identification was 1.94 mm (SD 1.45 mm).
• Manual annotations by clinicians showed a mean error of 1.32 mm (SD 1.08 mm), with minimum errors below 1.25 mm for specific landmarks.

Conclusions:

• Deep neural patchworks can accurately identify cephalometric landmarks on CT scans with mean errors below 2 mm.
• This automated method has the potential to significantly improve the workflow for cephalometric analysis in orthodontics and orthognathic surgery.
• The DNP algorithm's high precision and low training requirements make it a promising tool for clinical application.