Back to PLS Help
accessing to particular cluster
Posted on 09/28/08 12:08:41
Number of posts:
Dear Pls experts,
I have use your PLS package for E.R.FMRI analysis and it works perfect.
Now,I selected from the reported cluster for LV1 some particular clusters that I am mainly interested in and make a volume image just and only included those favourite clusters and set all the other clusters to zero.I thought using boot_result.compare might be a good clue to start with.Therefore,I took the bootstrap ratio of my first favourite cluster(Bootstrap ratio=7.5678) and wrote the following code for just bring out this cluster and set all the others to zero(I intend to write the same code for the other 7 desired cluster(off course with different Bootstrap ratio)):
Firstcluster_lv1=boot_result.compare(:,1)==7.4865;
LV1=Firstcluster_lv1.*boot_result.compare(:,1);
boot_result.compare(:,1)=LV1
but unfortunately the desired cluster was not found.I mean the first line which was:
Firstcluster_lv1=boot_result.compare(:,1)==7.4865;
can not find the cluster,so all the raw are zero(I expect a raw with one but there is not such a raw)
can any body let me know why is it like this?Shall I use the coordinate of the desired clusters rather than their bootstrap ratio?if yes,what is the best way of doing that.
Thanks in advance.
Sincerely,
Kianosh
Replies:
Untitled Post
I'm Online
jshen
Posted on 09/28/08 14:10:32
Number of posts: 291
- First, boot_result.compare does not have cluster information;
- Second, in order to get active BSR, the 1st row should be: Firstcluster_lv1=boot_result.compare(:,1)<-7.4865|boot_result.compare(:,1)>7.4865;
- third, follow the steps below to " ... make a volume image just and only included those favourite clusters and set all the other clusters to zero ...";
Step1: Open BSR view window, and change ratio to [-7.4865 7.4865];
Step2: Open cluster report window, and focus back to BSR window;
Step3: Select Cluster Mask under Report menu to display only the clustered voxels.
Step4: Click Save BSR result to IMG files under File menu;
You now have a volume image only including those clusters.
Untitled Post
Posted on 09/28/08 15:34:07
Number of posts:
Hello jimmy,
thank you for your prompt reply but I think that you misunderstood me.
For avoiding this,suppose that I am interested in just a cluster with bootstrap ratio=7.4958 be included in the volume image and all the other be zero.what you suggested will include all between [-7.4958 7.4958].right?
on the other hand all the reported clusters of the .imag files seem to be distorted since a cluster by [x,y,z] coordinate,will be shown up like [-x,y,z]?
Thanks,
Kianosh
Untitled Post
I'm Online
nlobaugh
Posted on 09/28/08 16:05:29
Number of posts: 229
Kianosh...
first of all.. be very careful about 'selecting' only clusters that are of particular to you from the PLS results. By doing so, you may mislead the reader about the extent and nature of the effects in your data. There is no problem with focusing on clusters "of interest" but I would not advise removing stable clusters from the presentation of the data.
The value in the cluster report is of the maximal bootstrap ratio in the cluster, and it's location, and you are unlikely to find this value useful for isolating individual clusters. It also likely that rounding in the report presentation contributed to your inability to find that voxel with your approach.
If you want to make separate images of the positive and negative clusters, you can do so by creating two cluster maps:
For the "Positive" set, set your Pos.Thresh to the value of interest (e.g., 3.0) and the Neg.Thresh value to the displayed Min. Ratio value
make & save the cluster report, mask your data, and save the img file
For the "Negative" set, reverse this: Neg.Thresh== -3.0, Pos.Thresh==Max. Ratio
make & save the cluster report, mask your data, and save the img file
nancy
Untitled Post
Posted on 09/28/08 16:41:49
Number of posts:
Hello nancy,
Off course I will not ignore the clusters with reliable and consistent activation. my purpose is just showing some predefined region only for visualization purpose.
Following your method,the strange point is that I have encountered with blank page on the cluster report while I set my Pos.Thresh to the value of interest (e.g., 7.49) and the Neg.Thresh value to the displayed Min.Actually I am interested in only one cluster with bootstrap ratio=7.49.This cluster was appeared while I set the negative and position threshold=+2.33(estimated by the software) but did not come up while I follow the way that you suggest me.
The second question is,according to your suggestion if I set my Pos.Thresh to the value of interest (e.g., 7.49) and the Neg.Thresh value to the displayed Min, it means that I am trying to keep those clusters with 7.49<bootstrap ratio<max positive bootstrap ratio,right?but actually that is not what I want, I only want a cluster with bootstrap ratio=7.49 to be shown up on the image file.if I set like the way that you suggest ,I might have a one with bootstrap ratio=8.76 beside the desired one?
Thanks
Untitled Post
I'm Online
nlobaugh
Posted on 09/28/08 18:54:26
Number of posts: 229
Hi..
there is no "cluster" with that value -
I repeat - you have focused on the value for the maximum bootstrap ratio in the cluster, as defined by the specific combination of options you selected for clustering, including bootstrap ratio thresholds. This is a single voxel, not a cluster.
all remaining voxels in that cluster will be in the range of the selected BSR threshold to max(or min) BSR for that LV.
we do not currently have (nor are we likely to provide) an easy method for manually extracting the voxels belonging to a single cluster - one reason for this is that the clusters are arbitrary, in the sense that if you change any of the parameters, the cluster results will also change (number of clusters, their sizes, cluster peak value & location)- this could possibly explain why as you change BSR values, you do not find a cluster that has the exact maximum that you see with 2.33. It could be that if you are sufficiently conservative (i.e, choose a BSR>=7), the remaining voxels no longer meet the chosen criteria for clustering.
hope this clears up your confusion
nancy
Login to reply to this topic.