EXPERIMENTAL VALIDATION OF FINITE ELEMENT MODELS OF THORACOLUMBAR VERTEBRAE

Abstract

Introduction: Computational modelling of the spine offers a particularly difficult challenge to analysts due to its complex structure and high level of functionality. Previous studies [Wijayathunga, 2008; Jones, 2007] have shown that finite element (FE) predictions of vertebral stiffness are highly sensitive to the applied boundary conditions and therefore validation requires careful matching between the experimental and simulated situation. The aim of this study was to develop and experimentally validate specimen specific FE models of porcine vertebrae in order to accurately predict the stiffness of single vertebra specimens.

METHOD: Nine single vertebra specimens were excised from the thoracolumbar region of two porcine spines. The specimens were mounted between two parallel PMMA housings and each specimen was imaged using a micro computed tomography (μCT) system (μCT80; Scanco Medical, Switzerland). In order to accurately match the experimental conditions, a radio-opaque marker was positioned on the specimen housing at the point of load application. The vertebrae were separated into two groups: a development set (set 1) consisting of three specimens and a validation set (set 2) of six specimens. Specimens from set 1 were used to establish the optimum method of conversion from image greyscale, to element material properties. The models in set 2 were used to assess the accuracy of the stiffness predictions for each model. The vertebrae were tested in a materials testing machine (AGS-10kNG; Shimadzu Corp., Japan) under axial compression and the stiffness for each specimen was calculated. The μCT data was imported into an image processing package (Scan IP, Simpleware, UK). The software enabled the images to be segmented and the vertebra, cement housings and position of load application to be identified. The segmented images were down-sampled to 1mm voxels, enabling a FE mesh to be generated (Scan FE; Simpleware, UK) based on direct voxel to element conversion. The Young’s modulus of each bone element was assigned, based on the greyscale of the corresponding image voxel. The PMMA housing plates were assigned homogeneous material properties (E = 2.45 GPa). Abaqus CAE 6.8 (Simula, Providence, Rhode Island, USA) was used for the processing and post-processing of all the models.

Results: The mean experimental stiffness was 4321 N/ mm (standard deviation = 415 N/mm). The optimum conversion factor was calculated for set 1, which yielded a root mean squared (RMS) percentage error of 7.5% when compared with the experimental stiffness. Using this optimised scale factor, FE models of specimens from set 2 were created. The predicted stiffnesses for set 2 were compared to the corresponding experimental values and yielded an RMS error of 10.1%.

Conclusion: The results indicate that specimen specific models can provide good agreement with the corresponding experimental specimen stiffness. In addition, the method employed in this study proved robust enough to be applied to vertebral tissue obtained from different animals of the same species. This method will now be developed to assess treatments for traumatic spinal injuries.