Date of Award



Department of Orthodontics

First Advisor

Joorok Park

First Committee Member

Heeyeon Suh

Second Committee Member

Heesoo Oh


Objectives: The aim of this study was to assess the accuracy and reliability of a fully automated landmark identification (ALI) system in automatically locating landmarks, in comparison to assessments made by human judges. Materials and Methods: We collected a total of 76 cone-beam computed tomography (CBCT) images. Following a calibration process, two human judges utilized In Vivo7 Software (Anatomage, San Jose, California) to locate 31 landmarks in the x, y, and z coordinate planes on the CBCT images. A ground truth dataset was created by averaging the landmark coordinates identified by the two human judges for each landmark. The accuracy of the ALI system was assessed by determining the mean absolute error (mm) for the x, y, and z coordinates, as well as the mean error distance (mm) between the human landmark identifications and those made by the ALI system. Additionally, the successful detection rate was calculated for each landmark. Results: In general, the ALI system demonstrated a similar level of success in landmark identification as the human judges. The average mean absolute error for all coordinates by the ALI system was 0.94 mm. Across all three coordinate planes, 94% of the landmarks had a mean absolute error of less than 3 mm. The mean error distance for all 31 landmarks was 1.99 ± 1.26 mm. When applied to the 31 landmarks across 76 CBCT images, the ALI system achieved a 91.85% success rate in detecting landmarks within a 2.5-mm error distance range. Conclusions: Overall, the ALI system yielded clinically acceptable mean error distances, with only a few exceptions. Notably, the ALI system demonstrated greater reliability than humans when identifying landmarks on the same image at different times. This study underscores the potential of ALI in assisting orthodontists with landmark identification on CBCT images. Introduction