Automation of model generation for numerical simulations based on computed tomography data

Numerical simulations of biomechanical problems by finite-element-analysis are common practice in the field of research and development. The required bone models are usually created from computed tomography (CT) data. A manual segmentation is very time consuming and prone to user dependent variability of results. Therefore, the goal of the present study was to generate a new automated method for computational model generation from CT data. Therefore, for every single CT slice the Hounsfield units along rays from every boundary voxel to each of the opposing voxels were evaluated assuming a circular cross section of the object and depending on a threshold given by the user. For each ray the CT numbers of consecutive voxels were compared and rising / falling edges (rising above / falling below the threshold) were counted. In case of an odd number of edge pairs every voxel between the first rising and the last falling edge and otherwise every second region was registered as “inside” locally for the particular ray. When a certain amount of “inside” cases – a percentage of belonging to any ray – was reached, the voxel was labeled as “bone” globally. The labels were automatically imported to the 3D reconstruction software Amira, ready to be processed into surfaces manually. The new algorithm has been applied to the CT data set of a human femur with implanted hip stem. A comparison of the femur to the results of a manual segmentation with regard to generated surface and volume showed that the diaphyseal bone structure can be mapped with deviations from -2.3 % to +4.9 %. Common segmentation techniques like thresholding and morphological operations did not lead to sufficient results. The presented algorithm proves as an appropriate method for supporting the transformation of CT data of human long bone into suitable solid body models applicable for subsequent numerical simulations.