Can CT image deblurring improve finite element predictions at the proximal femur?

Objective: Clinical application of computed tomography (CT) based finite element (FE) estimates of bone strength recently showed encouraging results [1]. However, model performance may be improved. In fact, in-vitro valida- tion studies showed systematic errors in strain prediction, especially where bone cortex is thin (i.e. the femoral neck) [2]. This work seeks to verify if, and to what extent, a CT deblurring algorithm restoring both geometry and intensity data in thin bone structures [3], can improve strain and failure load prediction accuracy of CT-based FE models of the prox- imal femur. Material and Methods: CT scans were acquired of 14 fresh-frozen human cadaveric femora. An estimate of the 3D Point Spread Function for each CT scan was used within a deconvolution solver to perform deblurring. Using the restored images, FE models of the proximal femur were generated [4]. Each femur was tested non- destructively in both stance and fall loading configurations to measure surface strains, and then loaded to failure in stance or fall. Deblurred FE predictions of strains and failure load were com- pared to experimental measurements, and FE predictions ob- tained from the original CT data (no deblurring). Results: An enhancement in strain prediction accuracy was obtained using deblurred FE models, with the Standard Error of Estimate reduced by 11 % with respect to reference FE models. Marked improvements at the femoral neck were achieved (e.g. peak error reduced by 38 %). Using deblurred models, the regression equation between FE-predicted and measured failure loads was characterized by a slope not sig- nificantly different from one, with R2 = 0.89, unchanged with respect to reference models. Absolute differences between estimated and measured failure loads were consistently re- duced by deblurring in stance (mean error 10 vs. 15 %) but not in fall (32 vs. 17 %). Conclusions: The proposed CT deblurring technique yielded moderate but significant improvements in femo- ral FE predictions, and can thus be seen as a first and worthwhile step in the improvement of CT-based FE models of the human femur.