ABSTRACT
With the growing demand for orthognathic surgery and other facial treatments, the accurate identification of anatomical landmarks has become crucial. Recent advancements have shifted towards using three-dimensional radiologic analysis instead of traditional two-dimensional methods, as it allows for more precise treatment planning, primarily relying on direct identification by clinicians. However, manual tracing can be time-consuming, mainly when dealing with a large number of patients. This study compared the accuracy and reliability of identifying anatomical landmarks using artificial intelligence (AI) and manual identification. Thirty patients over 19 years old who underwent pre-orthodontic and orthognathic surgery treatment and had pre-orthodontic three-dimensional radiologic scans were selected. Thirteen anatomical indicators were identified using both AI and manual methods. The landmarks were identified by AI and four experienced clinicians, and multiple ANOVA was performed to analyze the results. The study results revealed minimal significant differences between AI and manual tracing, with a maximum deviation of less than 2.83 mm. This indicates that utilizing AI to identify anatomical landmarks can be a reliable method in planning orthognathic surgery. Our findings suggest that using AI for anatomical landmark identification can enhance treatment accuracy and reliability, ultimately benefiting clinicians and patients.
ABSTRACT
BACKGROUND: Multi-dimensional facial imaging is increasingly used in hospital clinics. A digital twin of the face can be created by reconstructing three-dimensional (3D) facial images using facial scanners. Therefore, the reliability, strengths, and weaknesses of scanners should be investigated and approved; Methods: Images obtained from three facial scanners (RayFace, MegaGen, and Artec Eva) were compared with cone-beam computed tomography images as the standard. Surface discrepancies were measured and analyzed at 14 specific reference points; Results: All scanners used in this study achieved acceptable results, although only scanner 3 obtained preferable results. Each scanner exhibited weak and strong points because of differences in the scanning methods. Scanner 2 exhibited the best result on the left endocanthion; scanner 1 achieved the best result on the left exocanthion and left alare; and scanner 3 achieved the best result on the left exocanthion (both cheeks); Conclusions: These comparative analysis data can be used when creating digital twins through segmentation, selecting and merging data, or developing a new scanner to overcome all shortcomings.
