FBA / displaying significant results

fba

#1

Dear experts,

I did a FBA following the step by step guide provided in the [MRtrix3 documentation] (http://mrtrix.readthedocs.io/en/latest/fixel_based_analysis/ss_fibre_density_cross-section.html) with 9 athletes compared to 11 controls.

Everything worked out without any problem, fantastic job!

However, I am unsure how to proceed from step 23, especially which file I have to load in the fixel-plot tool to see significant results.

Running fixelcfestats with the -negative option I get the following output files:

-rw-rw-r-- 1 maudrich users   850864 Jun 20 14:51 "abs_effect.mif"
-rw-rw-r-- 1 maudrich users   850864 Jun 20 14:51 "beta0.mif"
-rw-rw-r-- 1 maudrich users   850864 Jun 20 14:51 "beta1.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 14:51 "cfe.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 14:51 "cfe_neg.mif"
-rw-rw-r-- 1 maudrich users  2495372 Jun 20 14:23 "directions.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 19:32 "fwe_pvalue.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 19:32 "fwe_pvalue_neg.mif"
-rw-rw-r-- 1 maudrich users 26839844 Jun 20 14:23 "index.mif"
-rw-rw-r-- 1 maudrich users    89464 Jun 20 19:32 "perm_dist_neg.txt"
-rw-rw-r-- 1 maudrich users    89491 Jun 20 19:32 "perm_dist.txt"
-rw-rw-r-- 1 maudrich users   850864 Jun 20 14:51 "std_dev.mif"
-rw-rw-r-- 1 maudrich users   850864 Jun 20 14:51 "std_effect.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 14:51 "tvalue.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 19:32 "uncorrected_pvalue.mif"
-rw-rw-r-- 1 maudrich users   850892 Jun 20 19:32 "uncorrected_pvalue_neg.mif"

I tried to load cfe.mif but don’t know with which file I have to threshold to display significant results and how to get the scaling right:

In short, I simply aim to get a similar output to that provided in this blog post under Fixel-based analysis.

Any help is greatly appreciated.

Thanks
Tom


#2

Hi Tom,

The image you’re looking for is fwe_pvalue.mif: This is the Family-Wise Error corrected p-value (well, actually the values in the image are of 1-p). So you want to load this image as the “threshold by” image, and setting the lower bound to 0.95 is equivalent to masking via p<0.05.

Any remaining fixels can then be coloured according to whatever contrast you want; cfe.mif contains the values resulting from the CFE statistical enhancement according to the equation presented in the paper, which may or may not be useful to you.

Cheers
Rob


#3

Thank you for the reply Rob!

Is there a possibility to use a different method to correct for multiple comparisons besides FWE and CFE, like for example FDR?


#4

The command also provides uncorrected p-values (the fraction of permutations where the enhanced statistic was greater than that of the default permutation for just that fixel, rather than using the maximal enhanced statistic across the entire image to build the null distribution). I’m not sure that we would actively pursue implementing alternative multiple comparison correction mechanisms unless there’s a strong justification for doing so.


#5

Hi! I ran some CFE analysis in order to compare FD,FC and FDC metrics between two groups and I’ve been struggling to understand the results. What file should I open in fixel plot ? What option should I choose in colour by and threshold by and what threshold should be applied?