Related Experiment Video
Updated: Jul 17, 2025
A Postoperative Evaluation Guideline for Computer-Assisted Reconstruction of the Mandible
Published on: January 28, 2020
Assessing the reliability of CBCT-based AI-generated STL files in diagnosing osseous changes of the mandibular
Kaan Orhan1, Alex Sanders2, Gürkan Ünsal3
1Faculty of Dentistry, Department of Dentomaxillofacial Radiology, Ankara University, Ankara, Turkey.
Objectives:
This study aims to evaluate the reliability of AI-generated STL files in diagnosing osseous changes of the mandibular condyle and compare them to a ground truth (GT) diagnosis made by six radiologists.
Methods:
A total of 432 retrospective CBCT images from four universities were evaluated by six dentomaxillofacial radiologists who identified osseous changes such as flattening, erosion, osteophyte formation, bifid condyle formation, and osteosclerosis. All images were evaluated by each radiologist blindly and recorded on a spreadsheet. All evaluations were compared and for the disagreements, a consensus meeting was held online to create a uniform GT diagnosis spreadsheet. A web-based dental AI software was used to generate STL files of the CBCT images, which were then evaluated by two dentomaxillofacial radiologists. The new observer, GT, was compared to this new STL file evaluation, and the interclass correlation (ICC) value was calculated for each pathology.
Results:
Out of the 864 condyles assessed, the ground truth diagnosis identified 372 cases of flattening, 185 cases of erosion, 70 cases of osteophyte formation, 117 cases of osteosclerosis, and 15 cases of bifid condyle formation. The ICC values for flattening, erosion, osteophyte formation, osteosclerosis, and bifid condyle formation were 1.000, 0.782, 1.000, 0.000, and 1.000, respectively, when comparing diagnoses made using STL files with the ground truth.
Conclusions:
AI-generated STL files are reliable in diagnosing bifid condyle formation, osteophyte formation, and flattening of the condyle. However, the diagnosis of osteosclerosis using AI-generated STL files is not reliable, and the accuracy of diagnosis is affected by the erosion grade.
