PhD Oral Exam - Behnaz Gheflati Feizabadi, Electrical and Computer Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Computer-assisted and robotic total knee arthroplasty (TKA) rely on accurate patient-specific anatomical models for surgical planning, alignment assessment, and intraoperative navigation. Statistical shape models (SSMs) provide a compact representation of anatomical variability, but conventional SSM coefficients are not directly related to clinically meaningful measurements, and the influence of patient-specific morphology on downstream surgical tasks remains incompletely understood. This thesis develops personalized statistical shape modelling methods for computer-assisted TKA, spanning anatomical reconstruction, alignment characterization, and registration reliability. First, a deep learning-based anatomically parameterized statistical shape model (DL-ANATSSM) was developed for femoral reconstruction. A nonlinear mapping between anatomical measurements and SSM coefficients was learned from synthetic shapes and fine-tuned using real patient data, reducing reconstruction error compared with a conventional linear anatomically parameterized model. The framework was then extended to tibial reconstruction by incorporating demographic information. Sex provided the largest demographic improvement, while body mass index contributed additional predictive information, and the fine-tuned model achieved the best reconstruction accuracy. The thesis next extended statistical shape modelling to whole-leg alignment analysis. A whole-leg SSM was used to distinguish intrinsic from apparent Coronal Plane Alignment of the Knee (CPAK) classification and to investigate the effect of knee flexion and coronal projection. Intrinsic and apparent classifications differed in 17 of 95 knees, demonstrating that imaging posture and projection can alter CPAK phenotype assignment, particularly near classification boundaries. Finally, the relationship between femoral morphology and surface-registration accuracy was evaluated using 3D-printed, synthetic, and in-silico experiments. Significant associations between SSM-derived shape variation and rotational registration errors demonstrated that registration reliability depends not only on algorithmic and acquisition factors but also on patient-specific bone geometry. Together, these studies show that SSMs can provide a unified framework for linking three-dimensional anatomy with clinically interpretable parameters, alignment phenotypes, and registration performance, supporting more personalized and anatomy-aware approaches to computer-assisted TKA.