University of British Columbia
Unveiling mandibular restoration : towards predicting functional and structural outcomes of mandibular reconstruction surgery via computational modeling
Abstract
dc:descriptionMandibular reconstruction is a central component of treatment for patients undergoing segmental mandibulectomy, particularly in the management of head and neck malignancies. Although virtual surgical planning has improved geometric reconstruction, long-term clinical success depends on broader outcomes, including restoration of mastication, achievement of donor-host bone union, and postoperative remodeling. This dissertation develops complementary computational frameworks to address these challenges through functional modeling, structural optimization, and predictive modeling. A dynamic biomechanical model of the reconstructed mandible is developed to evaluate mandibular kinematics, bite force, muscle activation, and condylar movement across representative defect types. The results show that reconstructed mandibles can recover substantial aspects of healthy masticatory behavior, while also demonstrating that defect extent and location materially influence postoperative function. This biomechanical framework is then extended toward patient-specific structural analysis and optimization. A workflow termed OsteoOpt, integrating virtual planning, mesh refinement, finite-element donor and fixation modeling, dynamic simulation, and Bayesian optimization, is proposed to improve the mechanical environment governing donor-host bone union. Across generic defect cases, optimized configurations increased cycle-averaged donor-mandible apposition by approximately 10–29 percentage points, and in patient-specific cases by approximately 9–26 percentage points over surgeon-implemented baselines. Longitudinal comparison against year-1 imaging in four cases yielded Dice overlap between 70% and 85% for predicted versus observed bone-formation patterns. Complementing these biomechanics-based components, a generative predictive framework, termed OsteoFlow, is introduced for forecasting postoperative bone remodeling from early postoperative computed tomography. Formulated as a conditional generative transport problem between Day-5 and Year-1 regions of interest, the method combines rectified-flow supervision, privileged-information guidance from registration, Lyapunov-guided trajectory distillation, and bone-focused endpoint supervision. Relative to the strongest non-teacher baseline, the proposed method achieved its most notable gain at the graft-host interface, reducing mid-slab bone mean absolute error by approximately 20%, while also improving overlap and structural similarity. Taken together, these contributions establish a unified computational perspective on mandibular reconstruction in which physics-based simulation and generative artificial intelligence play complementary roles. The dissertation shows that functional evaluation, structure-aware optimization, and longitudinal image-based prediction can be linked within a single planning-oriented framework for patient-specific surgical design and postoperative assessment.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Electrical and Computer Engineering
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Aftabi, Hamidreza
- Contributors dc:contributor
-
- Fels, Sidney
- Hodgson, Antony J.
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95150