In modelling applications such as custom-made: implants design is useful to
have a surface representation of the anatomy of bones rather than the voxe
l-based representation generated by tomography systems. A voxel-to-surface
conversion process is usually done by a 2D segmentation of the images stack
. However, other methods allow a direct 3D segmentation of the CT or MRI da
ta set. In the present work, two of these methods, namely the Standard Marc
hing Cube (SMC) and the Discretized Marching Cube (DMC) algorithms, were co
mpared in terms of local accuracy when used to reconstruct the geometry of
a human femur. The SMC method was found to be more accurate than the DMC me
thod. The SMC method was capable of reconstructing the inner and outer geom
etry of a human femur with a peak error lower than 0.9 mm and an average er
ror comparable to the pixel size (0.3 mm). However, the large number of tri
angles generated by the algorithm may limit its adaption in many modelling
applications. The peak error of the DMC algorithm was 1.6 mm but it produce
d similar to 70% less triangles than the SMC method. From the results of th
is study, it may be concluded that three dimensional segmentation algorithm
s are useful not only in visualisation applications but also in the creatio
n of geometry models. (C) 1999 Elsevier Science Ireland Ltd. All rights res
erved.