Subject: | | Re: Bayesian Contrasts DCM |
From: | | Ilenia Paparella <[log in to unmask]> |
Reply-To: | | [log in to unmask][log in to unmask]> wrote:
> Hi Jim, > > > > I’m sorry to bother you, but I just want to follow up on something you > said: > > > > “With FWE at .05, every voxel is statistically significant” > > > > Does “every voxel” mean ‘every voxel in the brain’ or ‘every voxel in the > cluster’. If it is the former, I don’t understand; I thought the point of > using FWE was that it would only identify clusters with a p-value below a > corrected alpha level… Any clarification you could provide would be > appreciated as I am still learning the intricacies of interpreting fMRI > results. > > > > Thanks, > > > > Josh > > > > *From:* SPM (Statistical Parametric Mapping) <[log in to unmask]> *On > Behalf Of *James Lee > *Sent:* Wednesday, March 16, 2022 2:10 PM > *To:* [log in to unmask] > *Subject:* Re: [SPM] FDRc/FDRp/FWE: interpretation question > > > > > *Attention:* This email originated outside of Penn State Health. Use > caution when clicking links or opening attachments. > > > > Patrick, > > > > Yes, when you are dealing with statistically significant clusters, you can > only say that the activation happened 'somewhere ' in that cluster. In this > analysis each voxel, on its own, is not statistically significant, so one > can't pretend that it is. > > > > This may NOT be very useful, because some of these clusters cover lots of > brain areas, so you are really limited in any conclusions you can draw > about which brain areas were contributing to the activation. > > > > With FWE at .05, every voxel is statistically significant, so every brain > area that contains a voxel is, in some sense, active, although I never make > any conclusions about single voxels, because it's too weak. > > > > There are lots of styles about reporting activation. Anatomic area, MNI > coordinates of the activation, and some measure of the strength of the > activation are all good. Percentage of voxels in different areas for a > single cluster is good in the sense that you are honestly reporting that > this huge cluster covered lots of areas. One of the no-no's is to have a > huge activation cluster covering many areas, and only mention the onr area > that contributes to your particular theory. > > > > Jim > > > > On Sat, Mar 12, 2022 at 3:19 AM Patrick < > [log in to unmask]> wrote: > > Dear SPM Experts, > > A while ago, I was attending a workshop and the speaker remarked that when > performing FDRc correction in SPM, the correct interpretation of each > cluster is that "there is at least one voxel within this cluster which is > statistically significant". Therefore, there is no way to really know which > brain regions (which the cluster covers) are activated per se. Does this > sound right or did I misunderstand something? > > Additionally, how does this interpretation differ when using FDRp or FWE > method of multiple comparison correction? If I see a large cluster covering > multiple brain regions, can I really not make any spatial localization > inference? > > Bonus: how should the re7cOs |
Date: | | Tue, 29 Mar 2022 16:17:03 +0200 |
Content-Type: | | text/plain |
Parts/Attachments: |
|
|
|
|
Dear Dr. Zeidman,
Yes, indeed we have a condition with light 1, another with light 2, and a condition with no light (or darkness). We don't have a condition with both lights. Participants are doing an oddball task in any of these 3 conditions. So far, in the design matrix I created for the DCM we have 3 columns: 1 with the onsets of the target sounds (sounds to be detected), another with blocks of light 1, and a third one with blocks of light 2. We thought to have an implicit baseline (darkness) which is basically anything happening "outside" light 1 and light 2, but maybe we need to add a new column with the blocks of darkness interleaving the blocks of lights?
Thank you in advance for your time and help.
Kind regards,
Ilenia
----- Mail original -----
De: "Zeidman, Peter" <[log in to unmask]>
À: [log in to unmask], [log in to unmask]
Envoyé: Lundi 28 Mars 2022 10:01:18
Objet: RE: Bayesian Contrasts DCM
Dear Ilenia
Taking a step back from the DCM, please can I clarify your experimental design? You mentioned that had two experimental conditions: light 1 and light 2. Did you also have conditions with no lights (i.e., baseline or null trials) and conditions with both lights, thereby forming a 2x2 design?
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Ilenia Paparella
Sent: 18 March 2022 12:33
To: [log in to unmask]
Subject: [SPM] Bayesian Contrasts DCM
⚠ Caution: External sender
Dear all,
I am trying to investigate with DCM the effects of two lights on the connections between 2 regions. That's what I did so far:
1. I have estimated a full model (full interconnection between the 2 regions and both lights modulating all the connections) and replicated it across subjects; 2. I checked the variance explained by the model and rejected subjects where the model was explaining less than 10% of the variance; 3. I run a 2nd level analysis (PEB) with no covariates separately for the B and the A matrix and then a PEB automatic search.
If I have only one of the 2 lights changing the baseline connection between my 2 VOIs, then I guess I don't have to do any further analysis. Am I right?
However, if I have both lights modulating the baseline connection I would like to compare the two modulations. In a previous email, Dr. Zeidman kindly told me that the best way to do it would be to have parameters in the B matrix that encode the difference between the 2 lights and then compare the evidence for models with those parameters switched on and off. But I'm not really sure how to do it.
Now my B matrix looks like this:
val(:,:,1) == [0 0; 0 0] input with no modulatory effects
val(:,:,2) == [0 1; 1 0] light 1 modulation on both connections
val(:,:,3) == [0 1; 1 0] light 2 modulation on both connections
How do I compute the difference between the 2 lights in the B matrix and compare the resulting models?
Thank you in advance for your time and help.
Best,
Ilenia
|
|
|
|