MVPA different labels TDT

1

Dear Martin,
I use the TDT software to analyse fMRI with a complex design, with a lot of categories (different modalities of the same categories).
I would like to analyse my scans for the same category in different modalities: for example, training the toolbox on the category A in a modality X vs other categories in modality X, and then test it on the same category A in modality Y vs the other categories in modality Y, to see wheter in a different modality, we can still decode the category. Is that clear?
Is it possible to do so with the toolbox?
Thanks a lot for this software and for your help,
Fabien

2

Hi Fabien,

I think what you are interested is known as cross-classification. Check out line 116 in decoding_tutorial.m for an example, which I reformatted for you:
% cfg = decoding_describe_data(cfg,{labelname_modXcatA labelname_modXcatOther labelname_modYcatA labelname_modYcatOther},[1 -1 1 -1],regressor_names,beta_loc,[1 1 2 2]);

Then, instead of make_design_cv, you would either use make_design_xclass or make_design_xclass_cv. The difference is that the former uses all runs, i.e. cross-classification within run is allowed, while the version with cv still does leave-one-run-out.

Hope that helps!
Martin

3

Hello,
Thanks for your fast answer, I will try it as soon as possible, but it looks like perfect!
Fabien

4

Hello Martin,

thanks for your toolbox! a quick question: If I have taining data and test data in two folders, how should I define beta_loc to pick this up? Now it seems it only accept one folder

5

Hi,

Instead of using beta_loc, you could use the template decoding_template_nobetas.m, where you manually pass the file names.

If this is too annoying, you could just run decoding_describe_data twice:
cfg1 = decoding_describe_data(cfg,{labelname1 labelname 2},[-1 1],regressor_names,beta_loc1);
cfg2 = decoding_describe_data(cfg,{labelname1 labelname2},[-1 1],regressor_names,beta_loc2);
and you could try merging the two with cfg = combine_designs(cfg1,cfg2);
This might not work, and then you would want to make both designs (e.g. with make_design_cv) and only then combine the designs, and later overwrite them with the design you actually want to use.

Hope this helps!
Martin