University of Twente Student Theses


Creating weight bearing 3D models of the lower limbs

Verdonschot, K.H.M. (2021) Creating weight bearing 3D models of the lower limbs.

[img] PDF
Abstract:Purpose: Lower limb malalignment is an important factor for young and active patients presenting knee osteoarthritis. Realignment osteotomy has proven to be highly effective as surgical treatment for those patients. An accurate preoperative plan is important for a successful outcome. The gold standard for preoperative osteotomy planning of the lower limbs requires weight-bearing whole leg radiographs (WLR). However, sagittal and transversal deformities can be overlooked on these 2D images. Additional 3D CT scans can provide this information but lack the weight-bearing aspect. The combination of 3D information with weight-bearing is especially useful in the care of patients presenting knee osteoarthritis with deformities in multiple planes. The aim of this research is to investigate the possibility of creating weight bearing 3D models of the lower limbs from a CT scan and a single WLR of the patient. Materials & Methods: Software was developed for manually aligning 3D models onto a single anteroposterior weight bearing WLR. This study included 30 patients with available CT scans. Digitally reconstructed (whole leg) radiographs (DRR) and anatomical 3D models were computed from these CT scans. Three raters performed manual registrations of the anatomical 3D models onto the DRRs using the software. A second method was developed for automation of the 3D-2D registration using an optimization algorithm. Anatomical 3D models of ten patients were registered using the semi-automatic algorithm with three different optimization algorithms. The registered 3D models were compared to the 3D models in original state. Errors were expressed in absolute distances and errors measured in the lower limb geometry: frontal hip-knee-ankle angle (HKA), sagittal HKA, joint line convergence angle (JLCA), and tibiofemoral rotation. Results: Mean registration error of the manual registration was highest in sagittal plane (6.10mm ± 4.47mm) compared to the anteroposterior plane (0.89mm ± 0.39mm). The angular error was highest for the sagittal HKA and tibiofemoral rotation, respectively 1.63° (± 1.28°) and 1.69° (± 1.33°), and lowest for frontal HKA and the joint line conversion angle, respectively 0.60° (± 0.60°) and 0.54° (± 0.64°). Mean registration error in the XYZ dimension was 30.10mm for the Genetic Algorithm (GA), 12.83mm for the multi objective GA, and 33.81mm for the surrogate algorithm. The angular errors were highest for the genetical algorithm (GA), with error of the frontal HKA being 2.32° (± 2.59°) and 1.57° (±1.18°) for the JLCA, while they were lowest when using a multi objective GA, 1.04° (±1.10°) for the frontal HKA and 0.93° (±0.77°) for the JLCA. Variation in results is high though. Conclusion: Manual registration of 3D models onto 2D DRRs provide accurate results in the antero-posterior plane, but results are less accurate in the sagittal plane. Semi-automatic registration using optimization algorithms is in the current form not accurate enough for clinical use.
Item Type:Essay (Master)
UMC Utrecht
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page