Hi everyone,

I'm new to functional connectivity analysis. So far, I've been working on task-based fMRI analyses with FEAT (both subject & group analyses) and I've been doing fine. I joined this project to learn some new techniques of fMRI analysis, but I've encountered some problems. I've downloaded the New_York_b Lite datasets and I've managed to run the fcon_1000 scripts without technical problems. I have all the output files, but I don't really know what to do with them now. Here is what I would like to do:

I'd like to look for the activated networks in the Broca's area in the r-fMRI scan and view them in the similar way to the way FEAT gives its reports, so I can easily see the activations, map them onto brain atlases, get the group mean and inter-subject differences. From what I understand, I have to create a Broca's seed using AFNI tools and use this seed in RSFC fcon script. That I can do. But what next? The problem is that I don't exactly understand what I get with the results of the RSFC analysis done by the script, and I can't find any relevant explanation as to what the resulting .nii file contains. Any information on what to do next to reach my goal will be appreciated.

Thanks,
Michal

Hi Michal,

Cool that you found your way to the datasets and through the scripts! Welcome :)

The results from the correlation analyses are single-subject 3D maps containging at every voxel the correlation of that voxel's time series with the time series of your seed. You can then combine these single-subject files into a group analyses.

I'd like to suggest the following papers as good startup reading:

Margulies, 2010, MAGMA
http://www.ncbi.nlm.nih.gov/pubmed/20972...
 
Cole, 2010, Frontiers in Systems Neuroscience 
http://www.ncbi.nlm.nih.gov/pubmed/20407... 

Margulies, 2007, NeuroImage
http://www.ncbi.nlm.nih.gov/pubmed/17604...

Hope this helps you on your way!
Maarten

Hi Maarten,

Thank you for the papers, I've read them and now I understand the concept of seed analysis, but I still have some questions regarding the technical side. How do I arrive at group reports with thresholded maps that I can compare? I browsed through the forum and I found a post which said that group analysis can be done using flameo. Could you give a simple explanation on how to do it? I ask, because the resulting files from RSFC are single 3D files and FEAT accepts 4D files or first-level folders (which I clearly do not have). Also, flameo asks for design files, so my question is how to prepare them? I tried to find relevant information on FSL website but I couldn't find anything which could help me with my dilemma. 

Thank you for your time,
Michal

Hi Michal,

glad we could help advance your understanding of the different measures. However, group analysis are not included in our scripts. What we typically do is just set up a group analysis using FSL FEAT, save it so it generates the .mat, .con and .grp files (typically don't worry about the fact that is complains not having any subjects) and then use the command line tools flameo (and easythresh) to run (and threshold) our analyses. You will have to manually concatenate the desired files into a 4D file and create a group mask for that. 

Not sure if you found this, but on the fsl website they have a section on how to setup group models. http://www.fmrib.ox.ac.uk/fsl/feat5/deta... 

Cheers,
Maarten

Hi Maarten,

Thank you very much for the information. I managed to run a group analysis without problems. I merged the files into a 4D volume and created a whole-brain mask, created the design files and successfully ran flameo. I am almost home, so I hope that this will be my final question. Now, I have the output from flameo(zstat1) which I would like to threshold, but I have problems with easythresh. I used standard values from FEAT (Z > 2.3, p<0.01) and what I got was one big yellow cluster all over the image. Is any additional normalization needed before using easythresh or am I doing something wrong? 

Thanks in advance,
Michal

Hello everyone,

it seems I'm really stuck at this point. I thought that the resulting file from RSFC analysis which is z-transformed is ready for thresholding , so I checked and I got the same results as in the group analysis I did before. I used standard values in easythresh (Z>2.3 and p<0.05) and I got the same all-yellow brain as a result. I also got NaNs in P and -log10(P) columns in cluster log when I used different values for Z. Does anyone know where the problem is? What should I do to get the clusters right?

Thanks,
Michal

Hi Michal,


The results will depend on the seed you used to calculate RSFC in the first place. What kind of seed did you use? How do the individual RSFC_Z maps look? Can you see a pattern there? You have to expect a similar pattern emerging in the group analysis.

The fact that you have nan P-values most likely has to do with a discrepancy between your mask and the data that go into the analysis. Carefully check that the analysis really includes only voxels that are present in all subjects. If I remember correctly you get this error if your mask is bigger than your actual data. (what we often do to mask is fslmaths 4dimage.nii.gz -abs -Tmin -bin outputmask.nii.gz)

The RSFC_Z maps are not zvalues, they are Fisher-Z transformed (a transformation to improve normality for correlation values).

Maarten

Hi Maarten,

Thank you very much for your help. Now I think I understand. I don't know why the whole time I've been thinking that _Z maps contain Z scores and now I know that is why I had problems with thresholding (which is due to fact that correlation simply cannot be over 1). Also thank you for the advice on masks, it worked and now the nan problem is gone. So now, can I actually threshold individual 3D images too, by giving coefficient values that might be statistically significant as Z input in easythresh?

As to the seed, I created a seed on Broca's area using MNI coordinates from the paper I read on Broca's and Wernicke's areas, using 3dundump and MNI152_T1_3mm_brain template. Is it the right way? I looked through the individual 3D maps from my RSFC anlysis and it seems that the biggest coefficient values are about 0.6 to 0.65. 

Thanks,
Michal

Hi Michal,

glad it worked out. I'm not sure I fully understand what you mean by thresholding individual 3D images, but if you just want to threshold an image for all voxels above a certain Z value, then just use 'fslmaths input.nii.gz -thr NUMBER ouput.nii.gz'. I would not use easythresh, as that also implements correction for multiple comparisons and takes into account the cluster size.

Yep, 3dundump is a good way to create a seed.

Keep up the good work!
Maarten

Hi Maarten,

Thank you very much for all your help. It works now, and I'm ready to do my study. 

Cheers,
Michal