Introduction: Virtual surgical planning is often used to prepare mandibular segment resection with subsequent reconstruction. Patient-specific cutting guides translate the planned resection and reconstruction osteotomies to the surgery. The accuracy of computer-assisted surgery is currently evaluated by heterogeneous methodologies for post-operative imaging, segmentation, registration, and accuracy measurements. Objective: This thesis aims to develop an objective, reproducible and insightful evaluation methodology. The evaluation methodology will compare planned osteotomies with actual osteotomies in segmental mandibular resection and reconstruction. The designed methodology will be applied in a retrospective study. Method: Actual osteotomies were defined by a plane fitted through manually defined points on the post-operative imaging. The actual osteotomies were aligned to the pre-operative mandible or fibula model. Distance and angular deviation were measured between the planned and actual osteotomies. Resection osteotomy distance deviation was defined as the distance between the centre of mass of the actual and planned intersection of the pre-operative mandible model and the osteotomies. The maximum distance between the intersections was also measured. Reconstruction osteotomy distance deviation was defined as the length difference between the planned and actual fibula segments. Angular deviation of the resection and reconstruction osteotomies were defined by two angular differences based on the saw slot, i.e. the angle across the saw slot (x-axis) and the angle through the saw slot (y-axis). Results: A semi-automatic novel methodology was developed. The intra-observer variation of the osteotomy localisation was ± 0.4 mm for distance deviation and ± 2.1° for the angular deviation. The inter-observer variation was ± 0.8 mm for the distance deviation and ± 2.4° for the angular deviation. Sixteen patients were included in the retrospective study. For the resection osteotomies, the absolute average distance deviation was 2.1 ± 1.9 mm for the centre of mass and 3.1 ± 2.3 mm for the maximum distance. The fibular segments differed by 2.4 ± 2.5 mm in length. Angular deviations around the x-axis were 3.7 ± 3.4° for resection and 6.9 ± 7.1° for reconstruction osteotomies, and deviations around the y-axis were 5.7 ± 5.8°and 9.1 ± 11.4°, respectively. Conclusion: The evaluation methodology provides guidelines for post-operative imaging, segmentation, osteotomy localisation, registration, and osteotomy comparison. The difference in distance deviated was within an absolute average of 3 mm. The angular deviation was significantly larger for the reconstruction osteotomies than for resection osteotomies, requiring further research.

Introduction: 3D imaging has become the standard for orthognathic surgery planning. Manual segmentation and landmark localization are the foundation for assessing deformities and planning of the corrective surgery. However, in 3D imaging, these manual procedures are complex and time consuming. Deep learning has emerged as a promising solution to automate this workflow. Objective: To develop an automated segmentation and landmark localization workflow to reduce time and efforted required for orthognathic surgery planning using CBCT. Methods: The dataset consisted of 57 presurgical Cone Beam CT (CBCT) scans. Manual segmentations were created for the mandible, maxilla, skin, mandibular canals (MC) left and right. Additionally, 43 landmarks (13 mandibular, 10 maxillary, 10 dental, and 10 surgical) were manually annotated. Automated landmark localization was approached as a segmentation task, with spherical segmentation of adjacent tissue with radii of 4mm, 5mm, or 6mm around landmarks. The dataset was split into 64% train, 16% validation 20% testset. Seven nnU-Net models were trained: one for segmentation and six (two models for each radius configuration) for landmark localization. Landmark positions were determined by calculating the center of mass of predicted segmentations. Performance was evaluated by comparing results to manual ground truth segmentations and landmark locations. Results: The nnU-Net was successfully trained to identify five segmentations and 43 landmarks. The overall median [Q1 – Q3] volumetric dice coefficient (vDSC) was 0.91 [0.79 - 0.96] for the segmentation model. The overall median radial error (MRE) for the landmarking models using 4mm, 5mm and 6mm spherical segmentations (each based on two models) were 0.98 [0.58 - 1.60] mm, 1.03 [0.63 - 1.67] mm, and 1.08 [0.66 - 1.82] mm, respectively. The successful detection rate below the clinical acceptability threshold of < 2mm ranged from 78.6% to 81.6%. The inference workflow required 17.7 minutes per patient on average. Conclusion: The nnU-Net achieved accurate craniomaxillofacial (CMF) anatomy segmentations and precise localization of landmarks, maintaining a clinical acceptability level with an error margin of less than 2mm. This fully automated workflow has the potential to enhance the efficiency of CMF planning.