Known Limitations

Slow permutations of Brunner Munzel tests for small groups

Voxels are used to split behavioral scores in two groups. If one of the groups (i.e., lesioned) has less than N=9, the Brunner Munzel tests are not valid and require permutation based thresholding (see Medina 2010). Currently, our fast method (BMfast) is not enabled to run permutations itself. This means that if your sample is N<90, you may be forced to use the slower "BM" method because there will be voxels affected in less then 9 subjects (with a a default minSubjectPerVoxel='10%'). LESYMAP will automatically tell you if you need to use the slow BM method. We may implement fast compiled permutations in later releases.
This is not anymore a limitation, we have implemented the fast BM permutations in v0.0.0.9007.

Difference with MRIcron's NPM approach

When performing permutation based multiple comparison correction with Brunner-Munzel tests, LESYMAP tracks the statistical score, while NPM tracks zscores. Typically the top BM score is also the top zscore, but this is not always guaranteed; zscores depend on degrees of freedom which vary from voxel to voxel. Despite the different approach, we have not observed major discrepancies between NPM and LESYMAP when using method='BMfast', multipleComparison='FWERperm'. However, it's worth knowing that the approach of the two software is slightly different.