help > Help interpreting VLSM NPM permutation results
Showing 1-2 of 2 posts
Nov 19, 2019 01:11 PM | marcelodelboni
Help interpreting VLSM NPM permutation results
Dear all,
We're conducting an exploratory binary images, binary groups VLSM analysis in NPM (2012) with ~60 patients with brain lesions divided into two groups: with a specific clinical deficit (coded as 0 in .val file) and without that deficit / "healthy" group (coded as 1 in the .val file). Only tested voxels damaged in at least 10%, set permutations = 4000, aiming at p<0.05 and run the stats. Could you please give us an advice on how to interpret the results below?
L Range -1.522...4.563
L+ FDR Z 0.050=2.21332455, 0.01=9.20000000
L- FDR Z 0.050=9.20000000, 0.01=9.20000000
L+: permutationFWE , 0.050=-0.00024465, 0.025=-0.00008664, 0.01=-0.00002900
L-: permutationFWE , 0.050=0.00038112, 0.025=0.00017590, 0.01=0.00007933
The FDR corrected results are great; when we display them using the threshold = 2.2133 we have nice results, compatible with current literature.
However, we don't understand the L+ permutation-FWER corrected results.
Why are Z values negative? Since they're negative we thought maybe these show the inverse "correlation" -i.e., brain areas more lesioned in the 1 group ("healthy")? How do we look at areas more lesioned in the 0, deficit group? Should we look at the L- tail? But why would we look at the L+ tail in FDR and the L- tail in permutation-corrected results?
Thanks a lot in advance for your help!
Marcelo
We're conducting an exploratory binary images, binary groups VLSM analysis in NPM (2012) with ~60 patients with brain lesions divided into two groups: with a specific clinical deficit (coded as 0 in .val file) and without that deficit / "healthy" group (coded as 1 in the .val file). Only tested voxels damaged in at least 10%, set permutations = 4000, aiming at p<0.05 and run the stats. Could you please give us an advice on how to interpret the results below?
L Range -1.522...4.563
L+ FDR Z 0.050=2.21332455, 0.01=9.20000000
L- FDR Z 0.050=9.20000000, 0.01=9.20000000
L+: permutationFWE , 0.050=-0.00024465, 0.025=-0.00008664, 0.01=-0.00002900
L-: permutationFWE , 0.050=0.00038112, 0.025=0.00017590, 0.01=0.00007933
The FDR corrected results are great; when we display them using the threshold = 2.2133 we have nice results, compatible with current literature.
However, we don't understand the L+ permutation-FWER corrected results.
Why are Z values negative? Since they're negative we thought maybe these show the inverse "correlation" -i.e., brain areas more lesioned in the 1 group ("healthy")? How do we look at areas more lesioned in the 0, deficit group? Should we look at the L- tail? But why would we look at the L+ tail in FDR and the L- tail in permutation-corrected results?
Thanks a lot in advance for your help!
Marcelo
Nov 19, 2019 03:11 PM | Chris Rorden
RE: Help interpreting VLSM NPM permutation results
I would heed the warning generated by NPM to upgrade to NiiStat.
https://github.com/neurolabusc/NiiStat
and
https://www.nitrc.org/plugins/mwiki/index.php/niistat:MainPage
All the source code for NPM is online on Github, but I have not looked at it for ~7 years, so am not well placed to answer you question.
https://github.com/neurolabusc/NiiStat
and
https://www.nitrc.org/plugins/mwiki/index.php/niistat:MainPage
All the source code for NPM is online on Github, but I have not looked at it for ~7 years, so am not well placed to answer you question.