tag:blogger.com,1999:blog-1405968300258104460.post2593750524409782546..comments2024-03-27T01:18:24.341-04:00Comments on Andy's Brain Blog: Psychophysiological Interactions ExplainedAndrew Jahnhttp://www.blogger.com/profile/16435706598096921650noreply@blogger.comBlogger17125tag:blogger.com,1999:blog-1405968300258104460.post-82975491818457482782019-01-26T01:01:50.867-05:002019-01-26T01:01:50.867-05:00Hi, Andy
Thank you for helpful information of SPM...Hi, Andy<br /><br />Thank you for helpful information of SPM analysis. I'm a beginner of SPM analysis, and have practiced using SPM12 and example data sets. I try to do PPI analysis, but faced to problem concerning contrast manager. Then, I would like to ask you for some helpful instructions. In SPM12 mannual, P.337, 36.5, PPI analysis, we can see the contrast "[2 1 -1; 3 1 1] for "stationary", "no-attention", and "attention" conditions. I cannot understand how the values in this contrast are calculated. I very appreciate you, if you kindly give us information.<br /><br />Sincerely yours,<br /><br />Takahiro Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-13988683871441366452018-05-15T17:01:03.192-04:002018-05-15T17:01:03.192-04:00Never mind - my regressors were wrong! Thanks Never mind - my regressors were wrong! Thanks Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-86186600895552345382018-05-11T11:58:30.537-04:002018-05-11T11:58:30.537-04:00Thanks so much! I seem to be having collinearity ...Thanks so much! I seem to be having collinearity issues popping up with so many regressors (each condition only appears for 16 seconds twice per run so most of the time points just have 0s). Is there some solution to this that you know of? If I ran a ridge regression perhaps?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-64578735947042389852018-05-09T15:02:58.709-04:002018-05-09T15:02:58.709-04:00Hi Andy,
Thanks for your very helpful blogs.
I ha...Hi Andy,<br /><br />Thanks for your very helpful blogs.<br />I had a very specific question related to gPPI (perhaps this isn't the best post to ask on but wasn't sure where I should!). I want to investigate patterns of functional connectivity between two different networks (let's say a social network A and a non-social network B). I can functionally localise these two networks using localiser tasks. Now I want to find whether regions in network A are functionally correlated with regions in network B. The way I've been doing this till now is to identify seed regions (let's say I take regions from network A as seed regions) and then see if they are functionally correlated with any other region in the brain, and then use a SVC in SPM to see if these regions are part of network B. However, are you aware of any analysis/toolbox that can allow me to use the functionally localised regions within network B as my search space instead of using the whole brain or SVC. Or more specifically can I use a seed to voxel based connectivity analysis where my voxels are only those which are functionally localised part of network B?<br /><br />I hope this ramble is more than a ramble and makes some sense! Any help would be very much appreciated! <br /><br />Thanks so much!<br />IsraAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-40739179802516700732018-05-06T17:08:42.479-04:002018-05-06T17:08:42.479-04:00Hi Heather,
Yes, this is a simple PPI analysis; t...Hi Heather,<br /><br />Yes, this is a simple PPI analysis; the field has moved on to gPPI, and I intend to get around to discussing it at some point. And yes, you should have an interaction regressor for each of your conditions, as well as modeling out the effect of each task.<br /><br />-AndyAndrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-30047624322720224372018-04-27T09:14:40.152-04:002018-04-27T09:14:40.152-04:00Hi Andy,
Am I right in thinking this is an exampl...Hi Andy,<br /><br />Am I right in thinking this is an example of how to conduct simple PPI and not generalized PPI? It seems the field has moved on to gPPI. With that methodology, if I have 9 conditions, even if I am only interested in the interaction of 2 conditions for a particular analysis, do you know if I am right in thinking that I should have an interaction regressor for each of my separate conditions?<br /><br />Thanks!<br />HeatherAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-51212064141113908232017-05-09T13:46:35.772-04:002017-05-09T13:46:35.772-04:00Hello,
You won't extract the time course from...Hello,<br /><br />You won't extract the time course from the GLM; the time course is extracted from the preprocessed data that goes into the GLM. For example, if smoothing was your last preprocessing step, then extract the time course from the smoothed data. Then enter that time course as a regressor into your GLM along with your interaction term and the other conditions from your "complex" GLM.<br /><br />Best,<br /><br />-AndyAndrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-82805335433983643982017-05-09T06:47:17.413-04:002017-05-09T06:47:17.413-04:00Hi Andy,
first of all thank you very much for you...Hi Andy,<br /><br />first of all thank you very much for your great blog and YouToube channel - they are very helpful for me as an fMRI newbie!!<br />I have a question regarding extracting the time series of the seed region in the PPI. I have already selected my seed region I am interested in and, now, I want to extract the time series of that region. But I am quite unsure what GLM I have to use. I usually use a very complex GLM for my first level analysis with lots of different conditions. I used this GLM to get the contrast of interest, i.e. my seed region. Do I have to use this complex GLM again to extract the time series or do I have to create a new GLM including just the conditions I am interested in? (I hope, it becomes clear what my problem is:)<br /><br />Thank you!<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-7569886382049900772017-04-14T14:26:41.940-04:002017-04-14T14:26:41.940-04:00Hey Po,
You'll want to capture the shape of t...Hey Po,<br /><br />You'll want to capture the shape of the entire HRF, which usually goes back to baseline after about 12 seconds. So, assuming you have a TR of 2s, you'd want 7 time points to capture the entire window (2*7 = 14 seconds).<br /><br />I notice that you're using the BLOCK5 basis function, with a 1 second duration; is that to reflect the actual duration of the events in your experiment (i.e., they lasted for 1 second)? If so, check your output over a longer time window - say, 30 seconds - and see how long it takes for the HRF of a 1-second event to go back to baseline, and use that time window to generate your HRF.<br /><br />Best,<br /><br />-AndyAndrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-74043954759846458292017-04-13T17:13:36.684-04:002017-04-13T17:13:36.684-04:00Hi Andy,
Thank you for your prompt reply!
But I ...Hi Andy,<br />Thank you for your prompt reply! <br /> But I think the boxcar function can't match my fMRI data with an event-related design. The Rmodel 'BLOCK5(1,1)' I used in the regression is a basic function for an event-related fMRI design. Maybe the HRF function for my PPI can be generated in this way:<br /> 3dDeconvolve -polort -1 -nodata timepionts TR -num_stimts 1 -stim_times 1 '1D: 0' 'BLOCK5(1,1)' -x1D Ideal -x1D_stop<br />But I don't know what's the number of time points should be inputted to the option '-nodata'. Could you tell me how to decide the inputted number of option '-nodata' to generated a HRF can matches the event-related design.<br />Po<br /> <br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-537468067841968682017-04-12T18:12:08.282-04:002017-04-12T18:12:08.282-04:00Hi Po,
You're right; the waveform should refl...Hi Po,<br /><br />You're right; the waveform should reflect the basis function used in the model. This tutorial was made a while ago, and that was an oversight.<br /><br />If you want to create a waveform representing a 5-second boxcar function, then you could do so with 3dDeconvolve:<br /><br />3dDeconvolve -polort -1 -nodata 20 1.0 -num_stimts 1 -stim_times 1 '1D: 0' 'BLOCK(5,1)' -x1D Ideal -x1D_stop<br /><br />I'm using a 20-second window, but you may want to shorten it or lengthen it as you see fit.<br /><br />Best,<br /><br />-AndyAndrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-7452292812939544452017-04-12T05:32:06.631-04:002017-04-12T05:32:06.631-04:00Hi Andy,
The form of the HRF generated by &quo...Hi Andy,<br /> The form of the HRF generated by "waver -dt TR -GAM -inline 1@1 > GammaHR.1D " is like a block response.I think it is because the input Rmodel of your following regression analysis is 'BLOCK(20,1)'.<br /> I will do PPI analysis to my fMRI data with event-related design. And I think it will be better to choose a response function which is the same as the one used in the following GLM regression. How to generate a HRF for event-related fMRI if the Rmodel used is 'BLOCK5(1,1)' in the following GLM regression?<br />I will be appreciated with your answers!<br />Thank you!<br />Po Anonymoushttps://www.blogger.com/profile/04738101872101855450noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-30727481070017771462013-09-29T16:26:44.085-04:002013-09-29T16:26:44.085-04:00I looked over the paper and talked about it with G...I looked over the paper and talked about it with Gang Chen, who wrote the AFNI guide for doing PPIs, and we were unable to determine why the values were corrected to a baseline of zero. It was unclear from the paper why they did this; your explanation is a possible reason, but again that would be more like a linear contrast than an interaction (because all of condition B would be negative, and all of condition A would be positive, similar to a [1 -1] contrast vector). I may need some clarification from the SPM programmers for why they approach it that way.<br /><br />Best,<br /><br />-AndyAndrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-88488430601148551202013-09-16T12:02:36.615-04:002013-09-16T12:02:36.615-04:00Hi Andy,
Thanks for the response!
I don't th...Hi Andy,<br /><br />Thanks for the response!<br /><br />I don't think I was clear in my original post. By correcting the Seed_Neur.1D file to a minimum of zero, I meant shifting all values within the file by a certain amount i.e. identifying the most negative value in your file and adding its absolute value to all values in the file, so that the minimum would then be zero. <br /><br />The reason I bring this up is because in Karl Friston's original paper on ppi analysis, he mentions correcting the physiological component to a minimum of zero prior to creating the interaction term (p. 223 just above figure 3 - " ... the activity in V1 (corrected to a minimum of zero) ...")*.<br /><br />I had presumed this step was taken to avoid the situation I outlined in my original post, but I'm confused as to why this step isn't mentioned in AFNI's guide to ppi.<br /><br />Any clarification is most appreciated! :)<br /><br />Thanks again!<br /><br />*FRISTON, K. J., BUECHEL, C., FINK, G. R., MORRIS, J., ROLLS, E. & DOLAN, R. J. 1997. Psychophysiological and modulatory interactions in neuroimaging. Neuroimage, 6, 218-29.<br /><br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-68804981028285534952013-09-13T15:07:17.652-04:002013-09-13T15:07:17.652-04:00Hi Anonymous,
The interaction term is looking for...Hi Anonymous,<br /><br />The interaction term is looking for voxels where activity is dependent on the condition, which requires weighting positive and negative values for the condition differently; think of a simple [1 -1 -1 1] contrast vector looking for a nonlinear interaction between levels of two conditions. Eliminating all negative values in the neural time file would make it more similar to a linear contrast between the conditions.<br /><br /><br />Hope this helps, but let me know if you have any other questions!<br /><br />-Andy Andrew Jahnhttps://www.blogger.com/profile/16435706598096921650noreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-39731498263563938092013-09-11T17:42:28.460-04:002013-09-11T17:42:28.460-04:00*"Neural Time"*"Neural Time"Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-1405968300258104460.post-72814161314944513542013-09-11T17:40:33.346-04:002013-09-11T17:40:33.346-04:00Just wondering if you need to correct the Seed_Neu...Just wondering if you need to correct the Seed_Neur.1D file to a minimum of zero before multiplying it by the AvsBcoding.1D file to form the interaction?<br /><br />This would avoid the issue of multiplying a negative "Neutral Time" value by a negative condition descriptor (-1) and thus generating a positive value in the interaction file, when it should in fact be a negative value.<br /><br />Great blog, by the way!Anonymousnoreply@blogger.com