Hi Alfonso,

you are absolutely right -- indeed the reason was a different version of sub2ind.m.

Solved now.

Thank you very much for helping out!

Ekaterina.

Originally posted by Alfonso Nieto-Castanon:

Hi Ekaterina,

Not sure, but my guess would be that you may have some toolbox which is overloading the original Matlab function sub2ind.m and causing this rather unusual behavior (???). Could you perhaps run in Matlab's command-line the command:

which sub2ind

and let me know whether that points to something like .../toolbox/matlab/elmat/sub2ind.m or somewhere else?

Best

Alfonso

Originally posted by Ekaterina Pechenkova:

Hi Alfonso,

sorry for bombarding the forum, but it seems I figured out the issue. 

The non-parametric stats computation indeed silently crashed without producing the results. 

In conn_vproject.m, line 3190

            conn_randomise(X,y,c,m,opt,THR,THR_TYPE,SIDE,niters,simfilename,[],adj,[],groupingsamples);

In conn_randomise, line 255

        adj=struct('size',tsize','index',sub2ind(tsize,1+datatype(1,:),1+datatype(2,:),1+datatype(3,:)));

I get an error:

Error using horzcat
Dimensions of matrices being concatenated are not consistent.

Error in sub2ind (line 60)
    k = [1 cumprod(siz(1:end-1))];

It looks like tsize should be transposed both times, and the code works smoothly till the end after changing to:

255     adj=struct('size',tsize','index',sub2ind(tsize',1+datatype(1,:),1+datatype(2,:),1+datatype(3,:)));

Hopefully this doesn't mess up the stats, and I got nicely looking results from one of my projects (attached).

So for now I just wonder why it was such an exotic issue I ran into :)

Best,

Ekaterina.

 

Originally posted by Ekaterina Pechenkova:

Hi Alfonso,

thank you for the explanation! 

However, it seems not to be the end of the story for me: I do get a prompt for a number of permutations and local/remote computing and then I get a pop-up saying "Preparing data. Please wait...".  But when computing cluster mass or some other non-parametric stats this pop-up is quickly replaced by another one saying "Updating non-parametric statistics. Please wait..." and showing the progress bar and ETA. "Updating non-parametric statistics" and permutation info are also usually printed into the logfile and Matlab command window. Nothing like that here with TFCE. "Preparing the data..." window just disappears in a minute without throwing any error, and that's all.

Does this mean that for TFCE permutations proceed silently in the background? Or do they just abort/not start in my setup for any reason?

Thank you and sorry for bothering.

Ekaterina.

 

Originally posted by Alfonso Nieto-Castanon:

Hi Ekaterina,

Those results (where all of the p-value columns still have no numbers in them and simply show '---' symbols) simply mean that the TFCE voxel-level statistics have still not been computed. TFCE requires randomization analyses to be run in order to compute p-values associated with each TFCE value (which can take a while, depending on the size of your dataset), after those are finished those '---' columns will be filled with the actual p-values associated with those TFCE values and you will be able to threshold the results normally. When you select TFCE you should be getting a prompt that asks you how do you want those permutation/randomization analyses to be run, you can specify there whether they should be run remotely (e.g. on a HPC cluster) or in this computer. Either way, after that simply wait (do not cancel those analyses) and when they are finished you will get a "normal-looking" results figure.

Hope this helps

Alfonso

Originally posted by Ekaterina Pechenkova:

Hello,

I'm puzzled by suspicios results for the voxel-to-voxel analysis with TFCE. While the RFT or regular non-parametric stats (cluster mass) result in a couple of small significant clusters, the TFCE shows almost all the brain covered by huge significant clusters and indicates a threshold of TFCE>0.00 and no calculations of p-values been done (see screenshot attached). Setting a more conservative p-value changes nothing.

I observe the same thing for different voxel-to-voxel measures (ICC, ILC, MVPA) and in at least two different versions of Conn (20b, 22a), for two projects at hand and for different contrasts in these projects.

Before I didn't use TFCE for connectivity analysis and only had experience with TFCE in the context of TBSS (DTI) analysis in FSL which does not tend to produce anything like this. 

What am I doing wrong? Or does it mean that nothing is actually significant?

The results screen illustrating TFCE on Conn site looks very reasonable and different from mine: https://web.conn-toolbox.org/fmri-method...

Could anyone please comment?

Thank you!