Thesis
Comparative analysis of AI-based mandibular segmentation software for CBCT-derived 3D reconstruction of edentulous mandibles
University of Iowa
Master of Science (MS), University of Iowa
Spring 2026
DOI: 10.25820/etd.008380
Abstract
Statement of problem: Accurate jaw segmentation of cone beam computed tomography (CBCT) data is fundamental to three-dimensional (3D) reconstruction of the edentulous jaws in dental implant digital workflows. Artificial intelligence (AI)-based automated jaw segmentation has emerged as a promising alternative to manual methods; however, evidence regarding its performance remains limited.
Purpose: The objective of this study was to compare the three-dimensional reconstruction accuracy of four commercially available AI-based mandibular segmentation software programs by using clinically derived CBCT datasets and investigate whether software-specific differences influence the geometric fidelity and surface mesh characteristics of reconstructed mandibular models.
Materials and Methods: Forty de-identified CBCT datasets of completely edentulous mandibles were selected from an institutional database of real patients after screening 250 datasets, adhering to strict exclusion criteria. Mandibles were segmented and reconstructed by using four different AI-based software programs: BlueSky Plan (version 5, BlueSky Bio), RealGUIDE (3DIEMME SRL), CoDiagnostiX (Dental Wings GmbH), and Relu Creator (Relu BV). The segmented jaw models were exported as Standard Tessellation Language (STL) format files and imported into a 3D image analysis software program (Geomagic Control X; 3D Systems), where a best-fit global and fine registration algorithm was used to align all models. Global surface-to-surface deviations were quantified through root mean square (RMS) error assessments for the entire mandible and specifically for the anterior region mesial to the mental foramen. The linear distance differences at a midline cross-section (height and width) and between bilateral gonial angles were compared among the groups. Additional geometric factors such as volume, vertex counts, total surface area, and density (vertex count-to-total surface area ratio) were calculated and compared. Statistical analysis was conducted by a linear mixed-effects model with restricted maximum likelihood estimation. The level of statistical significance was set at α =.05 for all analyses. Furthermore, a qualitative assessment was performed by color deviation maps of all groups.
Results: The global surface-to-surface deviation values at both the whole jaw and anterior segments indicated that CoDiagnostiX exhibited significant different deviation in comparison with other software programs (P<.001). In terms of the linear distance values at the bone height and width in the anterior region and between intergonial angles, BlueSky Plan software program group showed significantly larger values (P=.002), whereas no statistically significant differences were found among other software program groups (P>.05). Additional geometric analysis demonstrated significant differences among programs in volume, vertex count, surface area, and vertex density ratio with BlueSky Plan showing greater volume and CoDiagnostiX exhibiting higher surface area and vertex density (P<.05). Qualitative color-map analysis confirmed these findings, showing highest agreement between RealGUIDE and ReLu Creator, systematic surface expansion with BlueSky Plan, and greater overall surface deviation with CoDiagnostiX.
Conclusions: Statistically significant differences were identified among the four commercially available AI-based automated mandibular segmentation software programs in terms of global surface deviation, linear dimensional measurements, and geometric mesh characteristics. CoDiagnostiX demonstrated greater global and anterior surface deviation with geometric features suggestive of reduced enclosed volume and increased surface irregularity. BlueSky Plan exhibited significantly larger anterior linear dimensions and a systematic outward displacement pattern. In contrast, RealGUIDE and ReLu Creator showed closer agreement in global deviation metrics and qualitative surface reconstruction analysis, indicating comparatively more consistent segmentation outcomes among these platforms.
Details
- Title: Subtitle
- Comparative analysis of AI-based mandibular segmentation software for CBCT-derived 3D reconstruction of edentulous mandibles
- Creators
- Austin Green
- Contributors
- Seok-Hwan Cho (Advisor)Tarek El. Kerdani (Committee Member)Steven Dorris (Committee Member)Du Hyeong Lee (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Oral Science (Prosthodontics)
- Date degree season
- Spring 2026
- DOI
- 10.25820/etd.008380
- Publisher
- University of Iowa
- Number of pages
- x, 29 pages
- Copyright
- Copyright 2026 Austin Green
- Language
- English
- Date submitted
- 04/23/2026
- Description illustrations
- illustrations, tables, graphs
- Description bibliographic
- Includes bibliographical references (pages 25-29).
- Public Abstract (ETD)
Virtual dental implant placement is often planned by three-dimensional (3D) digital models created from cone beam computed tomography (CBCT), which produces detailed images of the jaw. To create these 3D models, specialized software programs must separate the jawbone from surrounding structures in a process called segmentation. Recently, artificial intelligence (AI) has been introduced to automate this step. However, studies about how consistent or accurate different AI programs are, remain limited.
The present study compared four commercially available AI-based software programs that create 3D models of toothless lower jaws from CBCT scans. Forty patient scans were analyzed, and the resulting digital jaw models were compared for differences in shape, size, and surface detail. Some differences were found among the software programs, including small variations in overall surface shape and jaw measurements. However, all differences remained within ranges considered acceptable for clinical implant planning even though the surface characteristics and reconstructed areas of the 3D models varied depending on the segmentation software programs used.
- Academic Unit
- Oral Pathology, Radiology and Medicine
- Record Identifier
- 9985177075102771
Metrics
1 Record Views