Thanks Peter and Dr. Seigher. Your comments were very helpful
-----Original Message-----
From: Zeidman, Peter [mailto:[log in to unmask]]
Sent: Wednesday, July 23, 2014 6:01 AM
To: Seghier, Mohamed; [log in to unmask]; Kapse, Kushal Janardan
Subject: RE: [SPM] DCM, max#VOI's
Ah! Moh is entirely correct, I mis-read the code, the data is not changed. So compare away :-)
Best,
P.
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Mohamed Seghier
Sent: 23 July 2014 10:54
To: [log in to unmask]
Subject: Re: [SPM] DCM, max#VOI's
Absolutely… you can compare different models with different nmax (known as the number of “modes”). Please note that the dimensionality reduction does not change the data, it only operates on the priors (basically by introducing some dependencies between the parameters, hence reducing the rank of the prior covariance matrix).
You can have a look at the function “spm_large_dcm_reduce.m” as well…
Best,
Mohamed
On 22/07/2014 21:27, Zeidman, Peter wrote:
> Thanks for that Mohamed! And Kushal, if you wanted to check that it doesn't change the results for your specific data, you could always do a model comparison (Bayesian Model Selection) between a model with nmax=9 and nmax=8.
>
> Best,
> Peter.
>
> Best,
> Peter.
>
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
> On Behalf Of Kapse, Kushal Janardan
> Sent: 22 July 2014 14:34
> To: [log in to unmask]
> Subject: Re: [SPM] DCM, max#VOI's
>
> Dr. Seghier, thanks a lot for your reply. That was helpful
>
> -----Original Message-----
> From: Mohamed Seghier [mailto:[log in to unmask]]
> Sent: Tuesday, July 22, 2014 9:11 AM
> To: Kapse, Kushal Janardan
> Cc: [log in to unmask]
> Subject: Re: [SPM] DCM, max#VOI's
>
>
> Hi Kushal,
>
> For models with many nodes, DCM tries to reduce the dimensionality of the problem in an informed way. This is done by using prior constraints that bound the effective number of free parameters. If you want to switch it off, you can follow Peter’s advice (by setting DCM.options.nmax = n nodes). Note that the dimensionality reduction does not affect the estimation of your parameters in a negative way (though their posterior estimates may change with nmax). My point here is that the dimensionality reduction in DCM does not hurt the model inversion.
>
> For more details, see: http://www.ncbi.nlm.nih.gov/pubmed/23246991
> Seghier ML, Friston KJ (2013) Network discovery with large DCMs.
> Neuroimage 68:181-191.
>
> I hope this helps,
> Best,
>
> Mohamed
>
>
>
> On 17/07/2014 14:05, Kapse, Kushal Janardan wrote:
>> Hi Peter,
>>
>> Thanks for the reply. The dataset I am applying to is fMRI and not resting state.
>> Does that mean that dimensionality reduction affects the estimation and Ep.B, Ep.C values in negative way? Because Across 9 VOI's, I am planning to setup around 60 models, and then perform BPA to study which connections have highest strength. Does setting large model space affects the parameter estimates in negative way?
>>
>> Thanks a lot
>>
>>
>> -----Original Message-----
>> From: Zeidman, Peter [mailto:[log in to unmask]]
>> Sent: Thursday, July 17, 2014 5:35 AM
>> To: Kapse, Kushal Janardan; [log in to unmask]
>> Subject: RE: [SPM] DCM, max#VOI's
>>
>> Hi Kushal,
>> The number of ROIs you can have is just a matter of what's practical - there's no hard limit (if you're using a script).
>>
>> If you have more than nmax regions, then a dimensionality reduction is performed prior to estimation (by modifying the priors, it seems).
>>
>> If you prefer, you could try setting nmax to 9 so this doesn't occur (DCM.options.nmax=9), however you will need to be conservative with which connections you allow , or you could end up with more parameters than are sensible. You could also try using the new "Estimate (cross-spectra)" option in the GUI, if your data is resting state, which is specifically designed for large models (see "A DCM for resting state fMRI", Friston et al. Neuroimage 2014).
>>
>> Best,
>> Peter.
>>
>> -----Original Message-----
>> From: SPM (Statistical Parametric Mapping)
>> [mailto:[log in to unmask]] On Behalf Of Kushal Kapse
>> Sent: 16 July 2014 17:15
>> To: [log in to unmask]
>> Subject: [SPM] DCM, max#VOI's
>>
>> Hi SPM users,
>>
>> I have a confusion/question regarding DCM max# VOI's to be included
>> in the DCM model specification. I do know that in GUI, DCM allows 8
>> VOI's only. But there is a provision where i can change this nmax to
>> greater numer. I am using 9 VOI's to specify my network and i am
>> using the script 'dcm_spm8_batch.m' and editing it to specify my
>> models. I have following two questions about DCM nmax VOI's
>>
>> 1- As i am using script to run DCM analysis, do i need to worry about nmax as my VOI's are 9 and the GUI limit is 8 (which i assume also should apply to script)???
>>
>> 2- If there is a need to change the DCM code in a way that i can
>> increase nmax to 9 or above, which specific script do I modify?????.
>> The current script doesnt show any option for that
>>
>> DCM.options.nonlinear = 0;
>> DCM.options.two_state = 0;
>> DCM.options.stochastic = 0;
>> DCM.options.nograph = 1;
>>
>>
>> I do know someone earlier did post this within current or last year and i did try to search on the thread, but i wasnt able to get this info.
>>
>> If anyone can throw the light on this, that would be helpful.
>>
>> Thanks
>> Kushal
>>
|