Evaluation and minimization of geometric reconstruction errors in FEM models generated from CT-scan images

CAD reconstruction of anatomical regions from computerized tomography (CT) scans is a very common approach in orthopaedic biomechanics. The CAD model is discretized into finite volume sub-domains and finite element (FE) analyses are performed in order to predict the distribution of stresses generated by applied loads. However, quality and reliability of numerical results depend on the level of accuracy reached in the meshing process. This paper analyzes some critical parameters that may affect the overall efficiency of the CT-FEM transformation process: scan threshold range, object size, and complexity. An optimization procedure for minimizing geometric errors on size and shape of reconstructed objects is presented. Finally, accuracy of stress predictions is evaluated for FE models that include known amounts of geometric errors. Compression and bending loads are considered. Results show that geometric and stress errors rapidly decrease as the objects to be reconstructed become larger in size. Optimal threshold ranges can be identified clearly for both an epoxy-resin benchmark model and a real bone specimen cut from a human lumbar vertebra. This allows geometric errors to be reduced significantly