Why is the new option not mentioned in the SPM12 manual? Regards,AnnaFrom: [log in to unmask] To: [log in to unmask] Subject: RE: [SPM] Serial correlations in SPM12 (About FAST option) Date: Fri, 21 Nov 2014 09:48:56 +0000 Why is the new option not mentioned in the SPM manual? Regards,Anna Date: Sat, 11 Oct 2014 15:52:54 +0000 From: [log in to unmask] Subject: Re: [SPM] Serial correlations in SPM12 (About FAST option) To: [log in to unmask] Dear Takanori, function spm_cfg_mfx.m was written before the FAST option was implemented and hasn't been updated accordingly. As you probably figured out, a simple fix is to edit the 'cvi' field for this extra option: matlabbatch{1}.spm.stats.fmri_spec.cvi = 'FAST'; I'd be interested to hear about the results of your investigations! Best regards, Guillaume. From: Takanori Kochiyama <[log in to unmask]> Sent: 09 October 2014 09:00 To: Flandin, Guillaume Cc: [log in to unmask] Subject: Re: [SPM] Serial correlations in SPM12 (About FAST option) Dear Guillaume, Thank you for your kind reply! I understood that FAST option is a possible implementation of flexibile covariance component model. I have another question related to FAST option. I am trying to investigate the effects of modeling with-subject variance (FAST option in this case) on population inference using mixed effects model, but the covariance component model is automatically changed to AR(1) model during mfx estimation. In fact, the code in spm_cfg_mfx.m lines 141-147 is as follows: if strncmp('AR',SPMS{1}.xVi.form,2) matlabbatch{1}.spm.stats.fmri_spec.cvi = 'AR(1)'; else %matlabbatch{1}.spm.stats.fmri_spec.cvi = 'none'; disp('Forcing serial correlation modelling to be AR(1).'); matlabbatch{1}.spm.stats.fmri_spec.cvi = 'AR(1)'; end Is it a feature or a sort of bug ? Thank in advance for helping. Best regards, Takanori. --------- Takanori Kochiyama, Ph.D. Researcher, ATR Brain Activity Imaging Center, Kyoto, Japan 2014-10-07 2:07 GMT+09:00 Guillaume Flandin <[log in to unmask]>: Dear Takanori, there is indeed a new option in SPM12 for the modelling of serial correlation (in first level fMRI) that uses a dictionary of covariance components based upon exponential covariance functions. The relevant code is in spm_fmri_spm_ui.m lines 244-265. Estimation proceeds as usual with ReML. The aim is to better model the faster frequency ranges in short TR data. The dictionary of variance components defined here seems to work satisfactorily on some datasets we looked at but it is possible that it might have to be adjusted for datasets exhibiting a different spectral pattern. See this code as an illustration on how to plug more refined covariance components in the SPM framework. Best wishes, Guillaume. On 06/10/14 09:49, Takanori Kochiyama wrote: > Dear SPM experts, > > I have a question about serial correlations in SPM12. > In SPM12, I can select FAST model in addition to iid and AR(1) model. > > As FAST model has a very generalised form of exponential covariance > structures, > I would think that it is suitable for very FAST fMRI time-series such as > short-TR multiband fMRI. > Is it correct? > > Thank in advance for helping. > > Best regards, > Takanori. > > --------- > Takanori Kochiyama, Ph.D. > Researcher, ATR Brain Activity Imaging Center, Kyoto, Japan -- Guillaume Flandin, PhD Wellcome Trust Centre for Neuroimaging University College London 12 Queen Square London WC1N 3BG