ADDRESSING PARTIAL VOLUME ARTIFACTS WITH QUANTITATIVE COMPUTED TOMOGRAPHY-BASED FINITE ELEMENT MODELING OF THE HUMAN PROXIMAL TIBIA
Quantitative computed tomography (QCT) based finite element modeling (FE) has potential to clarify the role of subchondral bone stiffness in osteoarthritis. The limited spatial resolution of clinical CT systems, however, results in partial volume (PV) artifacts and low contrast between the cortical and trabecular bone, which adversely affect the accuracy of QCT-FE models. Using different cortical modeling and partial volume correction algorithms, the overall aim of this research was to improve the accuracy of QCT-FE predictions of stiffness at the proximal tibial subchondral surface. For Study #1, QCT-FE models of the human proximal tibia were developed by (1) separate modeling of cortical and trabecular bone (SM), and (2) continuum models (CM). QCT-FE models with SM and CM explained 76%-81% of the experimental stiffness variance with error ranging between 11.2% and 20.2%. SM did not offer any improvement relative to CM. The segmented cortical region indicated densities below the range reported for cortical bone, suggesting that cortical voxels were corrupted by PV artifacts. For Study #2, we corrected PV layers at the cortical bone using four different methods including: (1) Image Deblurring of all of the proximal tibia (IDA); (2) Image Deblurring of the cortical region (IDC); (3) Image Remapping (IR); and (4) Voxel Exclusion (VE). IDA resulted in low predictive accuracy with R2=50% and error of 76.4%. IDC explained 70% of the measured stiffness variance with 23.3% error. The IR approach resulted in an R2 of 81% with 10.6% error. VE resulted in the highest predictive accuracy with R2=84%, and 9.8% error. For Study #3, we investigated whether PV effects could be addressed by mapping bone’s elastic modulus (E) to mesh Gaussian points. Corresponding FE models using the Gauss-point method converged with larger elements when compared to the conventional method which assigned a single elastic modulus to each element (constant-E). The error at the converged mesh was similar for constant-E and Gauss-point; though, the Gauss-point method indicated this error with larger elements and less computation time (30 min vs 180 min). This research indicated that separate modeling of cortical and trabecular bone did not improve predictions of stiffness at the subchondral surface. However, this research did indicate that PV correction has potential to improve QCT-FE models of subchondral bone. These models may help to clarify the role of subchondral bone stiffness in knee OA pathogenesis with living people.
QCT imaging, Finite Element modeling, Proximal tibia, Stiffness, Partial volume artifacts
Master of Science (M.Sc.)