Hi Darren,
Your second point was indeed the culprit. In testing multiple variants of
our hypothesized model (allowing bidirectional connectivity between
sustained and transient regions), the backwards variant has no driving
input. Thanks much for your help, Darren and Klaas!
Best,
~M
*****************************************
Melina R. Uncapher, Ph.D.
Functional NeuroImaging of Memory Group
Center for the Neurobiology of Learning and Memory
Department of Neurobiology and Behavior
University of California at Irvine
http://fnim.bio.uci.edu
[log in to unmask]
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of Klaas Enno Stephan
Sent: Monday, June 18, 2007 4:59 PM
To: [log in to unmask]
Subject: Re: [SPM] DCM parameter estimation problem
Hi Darren,
Yes, you are absolutely right - if this is the network structure, you
would indeed expect that the MAP estimates for the intrinsic
connections S1<->S2, S1->T1 and S2->T2 are zero.
Melinda, is this how your connectivity structure looks like?
All the best
Klaas
At 00:15 19/06/2007, Darren G wrote:
>Dear Melina
>
>I wonder about the structure of your DCM. I assume you have
>driving, intrinsic and possibly
>modulatory connections. From your description below you say the
>DCM.A matrix values corresponding
>to the sustained -> transient connections are 0, but I wonder if the
>DCM.A values for the transient
>network connections are 0. My guess is they are not.
>
>I am probably wrong but here is my suggestion: I wonder if you have
>a situation in which you do not
>actually propagate or connect your driving inputs into your
>sustained network? As an example, see
>below (I am ignoring modulatory inputs if any) T means a node in the
>transient network and S in
>the sustained network.
>
>
> INPUT --> T1 <- S1
> | ^
> | |
> v v
> T2 <- S2
>
>In this case the driving input enters at T1 which is connected to
>T2, but S1 and S2 never see this
>input since their connections are only backwards. I would expect
>DCM.A(T1 > T2) to have some value,
>but DCM.A(S1 > T1), DCM.A(S2 > T2) and DCM.A (S1 <> S2) would have
>values of 0.
>
>As I said this may be incorrect and Klaas will correct me if this
>would not be expected, but I hope
>this helps.
>
>Darren
>
>==============Original message text===============
>On Mon, 18 Jun 2007 5:15:50 pm CDT Melina Uncapher wrote:
>
>Hi Klaas,
>
>Thanks for your reply. Unfortunately, I'd already checked all those points
>and none appeared to be the culprit. The estimation appears normal, with
>~10-12 steps, no error messages, manual specification, all data were in
>their appropriate spots in the matrices, etc. Even changing the precision
of
>the printed values made no difference to the values. In fact, if I remove
>other connections the values are estimated fine. Has this been encountered
>in SPM5?
>
>Thanks,
>~Melina
>
>*****************************************
>Melina R. Uncapher, Ph.D.
>Functional NeuroImaging of Memory Group
>Center for the Neurobiology of Learning and Memory
>Department of Neurobiology and Behavior
>University of California at Irvine
>[log in to unmask]" target="_blank">http:[log in to unmask]
>
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
>Behalf Of Klaas Enno Stephan
>Sent: Monday, June 18, 2007 2:35 PM
>To: [log in to unmask]
>Subject: Re: [SPM] DCM parameter estimation problem
>
>Dear Melina,
>
>This sounds odd, and I am not sure what the problem could be. Here
>are a few things that you may wish to check (unless you have already done
>so):
>
>1. How many steps does the estimation take? Does it converge after a
>single step? If so, have you double checked that you specified the
>driving inputs correctly? Also, do you get any warnings or errors
>messages during estimation?
>2. Are you sure that the values in DCM.A are really zero, or not
>just close to zero? You can check this by printing the values at
>higher precision, using something like the sprintf command in Matlab.
>3. Did you manually specify the DCM that you are trying to estimate,
>or was it created algorithmically? If the latter, have you double
>checked whether anything went wrong in transferring the timeseries,
>i.e. does DCM.xY.u look like it contains your time series?
>
>This is all I can think of at the moment. You may also want to try
>estimating the model with the code in SPM 5. Generally, I would
>always recommend to use SPM 5 for DCM. The code is much more robust
>than in SPM 2.
>
>Best wishes,
>Klaas
>
>
>
>At 18:57 13/06/2007, you wrote:
> >Hi Klaas,
> >
> >I'm a postdoc working with Mick Rugg in California, and I've an odd (and
> >hopefully quick) DCM question for you. In running my fourth set of DCMs,
>I'm
> >running into a problem I've not encountered before. I'm looking at how
> >regions that exhibit sustained activity interact with those showing
> >transient activity, but when both sets of regions are included in the
same
> >model, the parameters in the A matrix corresponding to the connection
> >strength from the sustained to the transient regions do not seem to get
> >estimated. The DCM.a is appropriate, but the DCM.A contains 0s in the
> >relevant cells (yet has appropriate values in the cells for the visual
> >region to the transient regions).
> >
> >I should qualify this by stating I'm still using SPM2, as all my prior
DCMs
> >have been estimated in SPM2. I've tried to run diagnostics by
>systematically
> >excluding regions or otherwise changing the A matrix, yet I find no rhyme
>or
> >reason to when the parameters are estimated and when they're not. Have
you
> >encountered this before?
> >
> >Any help would be greatly appreciated. Mick sends his best.
> >
> >Thanks in advance,
> >~Melina Uncapher
> >
> >*****************************************
> >Melina R. Uncapher, Ph.D.
> >Functional NeuroImaging of Memory Group
> >Center for the Neurobiology of Learning and Memory
> >Department of Neurobiology and Behavior
> >University of California at Irvine
> >http://fnim.bio.uci.edu>[log in to unmask]
> >
> >
>===========End of original message text===========
>
>
>
>--------------------------------
>Darren R. Gitelman, M.D.
>Department of Neurology
>710 N. Lake Shore Dr. #1122
>Chicago, IL 60611
>Voice: (312) 908-8614
>Fax: (312) 908-5073
>--------------------------------
|