Hi,

Wolf is right - the max coordinates are meaningless in this case and you should ignore them.

The COG coordinates make much more sense and they are good to report as they still summarise where the results are located and that is the whole point of reporting coordinates.

All the best,

Mark

On 29 May 2013, at 17:49, zhang mingxia <[log in to unmask]> wrote:

I summed them. Yes, there are three value, 0,1 and 2. I also thought there was no max location after the binarization. But I did see "MAX X (vox)     MAX Y (vox)     MAX Z (vox)" reported, so I wondered what this stood for. 

Maybe I should report " COG X (vox)     COG Y (vox)     COG Z (vox)" in the paper because there is no peak information in this analysis? Did you see researchers report the Center of gravity instead of peak coordinates in the paper? I always reported the peak coordinated before.

Mingxia

On Wed, May 29, 2013 at 2:43 AM, wolf zinke <[log in to unmask]> wrote:

Hi Mingxia,

I think that the peak location is quite meaningless if you combine two binary masks. First question, do you conjoin these maps by summing them or by multiplying them? If you sum them up, you will have values of 0, 1, and two, otherwise you will have 0 and 1.

Anyhow, there will be no peak that can be localized, because the cluster likely consists only of flat levels, so I think you should ignore the max location and instead use the Centre of gravity.

I hope this helps,
wolf

On 05/29/2013 11:33 AM, zhang mingxia wrote:

Hi Wolf,

Thanks so much for your patient explanation! Due to the time limitation and other reasons, I prefer to use the old method I mentioned before to do the conjunction analysis. I would try the easythresh next time.

So, again, I come back to the question I raised before: When two binary image summed, there should be only 0 and 2 in the conjunction image. But When I used "cluster -t 2 -i conjunction_image", there was "MAX X (vox)     MAX Y (vox)     MAX Z (vox)" reported. What did this (MAX X (vox)     MAX Y (vox)     MAX Z (vox)) mean in this situation? 

Thanks again.

Mingxia

On Sun, May 26, 2013 at 9:18 AM, wolf zinke <[log in to unmask]> wrote:

Hi,

1. I saw the usages of easythresh: one is "easythresh stats/zstat1 mask 2.3 0.01 example_func grot" and the other is "easythresh stats/zstat1 2.3 example_func grot". It seemed in both kinds of usage, there was only one zstat image. To my understanding, the conjunction analysis is to conjoin two zstat maps. For example, if I want to conjoin zstat1 and zstat2, how should I do?

This is why you need to get the easythresh_conj script. This modification requires two zstat maps.

2. There are <cluster_z_thresh>(2.3 in the example) and <cluster_prob_thresh>(0.01). What is the <cluster_prob_thresh>?

puh, that one goes a bit more into the depth of the theory behind cluster based thresholding. The cluster_z_thresh is an arbitrary threshold that serves the cluster forming, and cluster_prob_thresh is used for inference based on Gaussian Random Field, i.e. it provides information about clusters that show 'significant' effects. Yo will find some good discussion of this arbitrary z-threshold and statistical inference in this paper:
Stephen M. Smith, Thomas E. Nichols (2009). Threshold-free cluster enhancement: Addressing problems of smoothing, threshold dependence and localisation in cluster inference, NeuroImage, http://www.sciencedirect.com/science/article/pii/S1053811908002978

I hope this helps,
wolf

Thanks again!

Mingxia

On Sat, May 25, 2013 at 3:46 AM, wolf zinke <[log in to unmask]> wrote:

Hi,

Thomas Nichols modified easythresh in a way that it is applying a conjunction analysis using a minimum statistics approach. You can find the script on his webpages (http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/nichols/scripts/fsl/easythresh_conj.sh), as well as some description and references for this approach ( http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/nichols/research/conjunctions/).

Since this approach uses the statistical maps, it would be possible to determine peak locations in terms of high statistical measures (i.e. z values). So maybe you are better off usng Thom Nichols' script for your conjunction analysis, since it does not binaries your maps.

good luck,
wolf

On 05/25/2013 12:18 PM, zhang mingxia wrote:

Hi Mark,

Thanks for replying. Yes, I added two binary images which from the significant zstat map of two contrast. So I suppose voxels where the value equals 2 are the significant results for the two contrast. Then, I focused on the conjunction map where I have thresholded at 2. 

I raised the question because I thought there are only one value on the conjunction map, that is 2, so there should not be any peak. But when I "cluster -i conjunctionmap -t 2", I did see peak were reported as other normal non-binary maps. I wondered whether these peaks are "real" peak? How are these peaks calculated? Can I report these peak in the paper(I didn't meet the binary maps before. For the  non-binary maps, we always reported peaks coordinates in the paper.)?   

One more question: I heard easythresh was also for conjunction analysis. Would easythresh theoretically get the same results as I did (use zstat maps to binarise and add binary maps to get conjunction map)?  

Thanks again.

Mingxia

On Sat, May 25, 2013 at 1:49 AM, Mark Jenkinson <[log in to unmask]> wrote:

Dear Mingxia,

I am a bit confused by what you are doing here.
Are you adding together two binary images are then wanting to find peaks (i.e. voxels where the value equals 2)?

And I'm not sure what you mean by "the peak coordinates are reported (MAX X (vox)     MAX Y (vox)     MAX Z (vox))".  Do you mean that they are now in voxel coordinates and that is not what you want?  These columns do reflect the positions of the maximum values, and that is normally how they are labelled.

I can't really help more at this stage.
All the best,
        Mark

On 24 May 2013, at 09:44, zhang mingxia <[log in to unmask]> wrote:

> Dear FSL experts,
>
> I have done the conjunction analysis: binarise two images and add them to one image (con_image.nii.gz). Without conjunction analysis, we always report the peak coordinates. After the conjunction analysis, although I see the peak coordinates are reported (MAX X (vox)     MAX Y (vox)     MAX Z (vox)) when I threshold at 2 (cluster -t 2 -i con_image.nii.gz), I am curious how that peak calculated. Is that a real peak? Can I report that?
>
> Thanks in advance!
>
> Mingxia Zhang