In the BoneJ plug-in, I am running the Anisotropy tool to determine the fabric tensor of a parallelepiped sub-region of a vertebra. I think I would like to use a single sphere at the center of this volume of interest (VOI), but when I select “Single Sphere” as an option, “ NaN" are returned for the fabric eigenvalues and eigenvectors (and 1.000 is returned for DA and infinity is returned for tDA). It’s likely that I don’t understand what the “Single Sphere” option is doing. But more importantly (I think), there are two parameters that do not seem to be explained in the information provided online for BoneJ Anisotropy, : “radius” and “vector sampling”. Even in the articles cited online, there is no mention of these specific options. I am assuming that “radius” is the vector length used in the MIL algorithm and that “vector sampling” is the “arc” distance between each vector, but is this right? If you could please confirm or correct on these assumptions, I would really appreciate it. I have run the Anisotropy routine within BoneJ several times for the same VOI, and I get different results each time, and I am wondering if this variation, or at least some of this variation, is due to using values of radius and vector sampling that are ill-suited to our specimens. Thank you for providing this essential evaluation tool.

# BoneJ Anisotropy Options

**mdoube**#2

You are right, radius is the vector length. Vector sampling is the distance between sample points on each sampling vector. MIL works by detecting boundaries. A boundary is counted on a vector when subsequent sample points belong to different phases of the image. The default sampling is set to be conservative, a little less than half pixel spacing, to make sure that most boundaries along a vector are counted, at the expense of run time.

The NaN and infinity results sound like no boundaries were detected, or they were detected in a single plane. The instability you’ve observed may result from a small VOI relative to pixel spacing or texture coarseness, interacting with some random sampling features of the implementation.

**jloaiza**#4

Great! Thank you for clarifying these options and adding that information on the website.