Hi all I think Greg was making the valid point that: if you want to regress the full space of A and B out of your data, it is not correct to regress out A and then afterwards regress out B, unless A and B are orthogonal. To do it correctly you either need to combine [A B] into a single model to regress out of the data, or else regress A out of the data and also out of B, and then regress the new B out of the data.
Hope that helps, Steve. > On 10 Dec 2015, at 08:46, Maarten Mennes <[email protected]> wrote: > > Hi Matt, > > I'm not sure I follow your reasoning under 1). I would prefer a much richer > 'signal vs. noise' model as provided by ICA+FIX compared to aggressive > regression of (potentially GM contaminated) WM and CSF signal. This is also > what you indicate at the end of your answer: regressing out WM and CSF after > ICA+FIX doesn't seem to make much of a difference. > > With 3) I believe Greg is pointing to combining all regressors into one > model, ie. doing non-aggressive motion, WM, and CSF regression, by adding > those regressors to the ICA+FIX generated model. There I again don't see what > the benefit would be of adding very crude WM/CSF regressors over the more > fine-grained ones ICA+FIX is providing. But it seems you are working on a > manuscript that addresses some of these points. Looking forward to that then! > > Cheers, > Maarten > > > > On Thu, Dec 10, 2015 at 3:39 AM, Glasser, Matthew <[email protected] > <mailto:[email protected]>> wrote: > 1) Greg I covered this in my comments on your manuscript. For the benefit > of others: it is not the case that ICA+FIX would replace using WM and CSF > timecourses. The reason is that ICA+FIX regresses noise components out of > the data in a ³non aggressive² approach (computing betas for all > components and then subtracting only the data explained by the noise > components). This is different from how the movement parameters are > removed (where all of the variance explained by those timeseries is > aggressively regressed out). If you aggressively regress out WM and CSF > timecourses, this will remove all variance explained by them, even that > which is shared by the signal components and would not be removed in > ICA+FIX. It is important that your WM and CSF regressors are kept well > away from the grey matter, as you do not want to be doing GSR ³light² (or > any GSR/MGTR for that matter). I have investigated whether removing WM or > CSF regressors is helpful. For WM, it really doesn¹t make much of a > difference unless there are MR acquisition artifacts (e.g. some receive > coil elements stopped working during the scan). One of the reasons we > know there aren¹t global movement related effects being left in the HCP > data is that regressing WM out doesn¹t do much (if there were, one would > expect them to be picked up by the WM timeseries). CSF may be more > correlated with physiological noise, but it isn¹t a very clean regressor > for this (nor is WM). > > 2) Indeed there does appear to be global grey matter physiological noise > which we need to separate from global neural signal. We are working on > ways to do this (it is likely that we will want to investigate both > external approaches like regressing physiological regressors out and > internal approaches that attempt to separate physiological noise from the > data itself). It would certainly be helpful if the HCP¹s physiological > noise regressors were preprocessed for heart rate and respiratory end > tidal volume traces and those regressors were released publicly. > > 3) I don¹t understand this. > > Peace, > > Matt. > > On 12/9/15, 5:39 PM, "[email protected] > <mailto:[email protected]> on behalf of > Greg Burgess" <[email protected] > <mailto:[email protected]> on behalf of > [email protected] <mailto:[email protected]>> wrote: > > >Just a few comments: > > > >1) I agree with Maarten that FIX-denoising is effectively removing WM and > >CSF components from the FIX timeseries data. My understanding is that > >Matt Glasser has evaluated the incremental benefit for regressing the > >(average) WM and CSF timeseries and concluded that there is no additional > >benefit after FIX. I don¹t believe that anyone has tested whether > >additional WM and CSF components (such as implemented in CompCor) can > >remove noise variance above and beyond FIX, though, in theory, FIX should > >capture those as well. > > > >2) Some analyses that I have conducted suggest that FIX denoising may > >leave behind some proportion of physiological noise, especially that > >which is more globally-distributed across gray matter. I am hoping to > >investigate whether physiological regressors can remove that additional > >proportion. > > > >3) It is probably important to regress physiological regressors, motion > >regressors, and FIX noise regressors simultaneously from the timeseries. > >Otherwise, the fit of some regressors to the timeseries will be worsened > >by removing the other regressors from the timeseries. It might work to > >regress motion and noise ICs from the physio regressors, and then regress > >physio from the FIX timeseries, but I am not certain that it will (e.g., > >for RETROICOR-type regressors). > > > >4) If you decide to investigate physiological regressors, please use > >those release in the 900 subject packages, because the physiological > >measures in the 500 subject release had a timing bug. There will be > >additional information about the updated physiological measures coming in > >the near future. > > > >--Greg > > > >____________________________________________________________________ > >Greg Burgess, Ph.D. > >Staff Scientist, Human Connectome Project > >Washington University School of Medicine > >Department of Neuroscience > >Phone: 314-362-7864 <tel:314-362-7864> > >Email: [email protected] <mailto:[email protected]> > > > >> On Dec 9, 2015, at 4:40 PM, Maarten Mennes <[email protected] > >> <mailto:[email protected]>> > >>wrote: > >> > >> One could also argue that WM and CSF regression is not needed anymore > >>if the FIX denoising worked as intended... Given that FIX will look for > >>components that correlate with WM and CSF signal these signals are > >>already regressed out if components were properly identified. The same > >>is true for the physiological data. > >> > >> Note that this is different from a tool like ICA-AROMA which is not > >>trained to identify WM/CSF/physiological components, in this case extra > >>nuisance regression might effectively come in handy. But I don't see > >>this need in FIX-denoised data. > >> > >> Or am I missing something obvious? > >> > >> Cheers, > >> Maarten > >> > >> > >> On Wed, Dec 9, 2015 at 11:10 PM, Harms, Michael <[email protected] > >> <mailto:[email protected]>> > >>wrote: > >> > >> See inline below. > >> > >> -- > >> Michael Harms, Ph.D. > >> ----------------------------------------------------------- > >> Conte Center for the Neuroscience of Mental Disorders > >> Washington University School of Medicine > >> Department of Psychiatry, Box 8134 > >> 660 South Euclid Ave. Tel: 314-747-6173 <tel:314-747-6173> > >> St. Louis, MO 63110 Email: [email protected] <mailto:[email protected]> > >> > >> From: Joelle Zimmermann <[email protected] > >> <mailto:[email protected]>> > >> Date: Wednesday, December 9, 2015 3:08 PM > >> To: "Harms, Michael" <[email protected] <mailto:[email protected]>> > >> Cc: "[email protected] <mailto:[email protected]>" > >> <[email protected] <mailto:[email protected]>> > >> Subject: Re: [HCP-Users] FIX-denoised > >> > >> Hi Michael, > >> > >> Thanks for your help. I have a few more questions below.. > >> > >> Has an average timeseries for WM (and CSF) signal been already computed > >>by HCP? > >> Perhaps this is the "rfMRI_REST1_LR_WM.txt" and the > >>"rfMRI_REST1_LR_CSF.txt" in the FIX extended package? > >> Yes, you could use those. See the scripts in the RestingStateStats > >>folder in github if you want to know the details of how exactly those > >>were derived. But see next for some downstream files that you could use > >>instead that would save you a lot of methodological preparation. > >> > >> Is there a recommended way of regressing these out? > >> Take a look at the RestingStateStats/RestingStateStats.m code. In > >>particular, if starting from the "MPP" WM and CSF time courses (i.e., > >>from the non-cleaned data) you need to apply the same HP filter, and > >>regress out the motion parameters and the FIX-identified noise > >>components, so that you don¹t re-introduce noise related to those > >>operations into the cleaned data. And if you were to regress out WM and > >>CSF sequentially, then you need to do something similar to account for > >>the order there as well. It gets a bit complicated. :) > >> > >> Fortunately, looking at the RestingStateStats.m script, there should be > >>files "*_Cleaned{WM,CSF}tc.txt" that have already done all that for you. > >> In which case, all you need to do is use those simultaneously in a > >>regression to remove the noise space spanned by those two already > >>supplied "*_Cleaned{WM,CSF}tc.txt" files from the hp2000_clean.nii.gz > >>volume data or hp2000_clean.dtseries.nii CIFTI data. (See e.g., Lines > >>330-334 in RestingStateStats.m for what I mean by using them in a > >>"simultaneous" regression). > >> > >> I think I have that right, but Matt wrote that particular bit of code, > >>so hopefully he will correct if I've misstated anything. > >> > >> Related to my previous question: Has physiological data > >>(³rfMRI_REST1_LR_Physio_log²) already been regressed from the FIX > >>denoised data? > >> I cannot actually find a rfMRI Physio log file in the FIX-ed dataset, I > >>can only find this Physio log file in the minimally preprocessed > >>dataset. I suppose I can use that one? But I'd assume there should be > >>one in the FIX-ed dataset folder as well... > >> We haven't yet extended the RestingStateStats.m code to incorporate > >>regressors derived from the physio files, which is why the physio time > >>series isn't also part of the FIX-extended packages. If you want to try > >>regressing out physio as well, again make sure that you first regress > >>out of any physio regressors all the preceding filtering/regression > >>steps, so that you don't re-introduce any previously removed noise from > >>the cleaned data. > >> > >> > >> Thanks, > >> Joelle > >> > >> On Wed, Dec 9, 2015 at 2:43 PM, Harms, Michael <[email protected] > >> <mailto:[email protected]>> wrote: > >> > >> Hi, > >> See inline below. > >> > >> -- > >> Michael Harms, Ph.D. > >> ----------------------------------------------------------- > >> Conte Center for the Neuroscience of Mental Disorders > >> Washington University School of Medicine > >> Department of Psychiatry, Box 8134 > >> 660 South Euclid Ave. Tel: 314-747-6173 <tel:314-747-6173> > >> St. Louis, MO 63110 Email: [email protected] <mailto:[email protected]> > >> > >> From: Joelle Zimmermann <[email protected] > >> <mailto:[email protected]>> > >> Date: Wednesday, December 9, 2015 1:26 PM > >> To: "[email protected] <mailto:[email protected]>" > >> <[email protected] <mailto:[email protected]>> > >> Subject: [HCP-Users] FIX-denoised > >> > >> Hi everyone, > >> > >> I'm interested in using the FIX-denoised data, and am currently looking > >>at the extended package, as I'm interested in the volumetric data. > >> > >> I'm wondering whether the motion parameters have already been regressed > >>out from the FIX denoised rfMRI_REST1_LR_hp2000_clean.nii.gz? > >> YES > >> > >> I'm assuming yes, that the FIX denoised actually already deals with > >>this by separating the motion-related noise into a component and > >>filtering that out of the signal? I assume so based on the following > >>snippet from the manual describing FIX-ed data: > >> "As part of this cleanup, we also used 24 confound timeseries derived > >>from the motion estimation (the 6 rigid-body parameter timeseries, their > >>backwards-looking temporal derivatives, plus all 12 resulting regressors > >>squared ‹ Satterthwaite et al., 2013). The motion parameters have the > >>temporal highpass filtering applied to them and are then regressed out > >>of the data aggressively, as they are not expected to contain variance > >>of interest." > >> Am I correct? > >> YES > >> > >> Has white matter and/or cerebrospinal fluid been regressed already from > >>the FIX denoised fMRI timeseries? > >> NO > >> > >> Has physiological data (³rfMRI_REST1_LR_Physio_log²) already been > >>regressed from the FIX denoised data? > >> NO > >> > >> Any pointers would be much appreciated. > >> > >> Thanks, > >> Joelle > >> > >> > >> > >> > >> > >> > >> _______________________________________________ > >> HCP-Users mailing list > >> [email protected] <mailto:[email protected]> > >> http://lists.humanconnectome.org/mailman/listinfo/hcp-users > >> <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> > >> > >> > >> > >> > >> The materials in this message are private and may contain Protected > >>Healthcare Information or other information of a sensitive nature. If > >>you are not the intended recipient, be advised that any unauthorized > >>use, disclosure, copying or the taking of any action in reliance on the > >>contents of this information is strictly prohibited. If you have > >>received this email in error, please immediately notify the sender via > >>telephone or return mail. > >> > >> > >> > >> > >> > >> The materials in this message are private and may contain Protected > >>Healthcare Information or other information of a sensitive nature. If > >>you are not the intended recipient, be advised that any unauthorized > >>use, disclosure, copying or the taking of any action in reliance on the > >>contents of this information is strictly prohibited. If you have > >>received this email in error, please immediately notify the sender via > >>telephone or return mail. > >> > >> _______________________________________________ > >> HCP-Users mailing list > >> [email protected] <mailto:[email protected]> > >> http://lists.humanconnectome.org/mailman/listinfo/hcp-users > >> <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> > >> > >> > >> > >> > >> -- > >> Maarten Mennes, Ph.D. > >> Senior Researcher > >> Donders Institute for Brain, Cognition and Behaviour > >> Radboud University Nijmegen > >> Nijmegen > >> The Netherlands > >> > >> Google Scholar Author Link > >> _______________________________________________ > >> HCP-Users mailing list > >> [email protected] <mailto:[email protected]> > >> http://lists.humanconnectome.org/mailman/listinfo/hcp-users > >> <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> > >> > > > > > >_______________________________________________ > >HCP-Users mailing list > >[email protected] <mailto:[email protected]> > >http://lists.humanconnectome.org/mailman/listinfo/hcp-users > ><http://lists.humanconnectome.org/mailman/listinfo/hcp-users> > > > ________________________________ > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you are > not the intended recipient, be advised that any unauthorized use, disclosure, > copying or the taking of any action in reliance on the contents of this > information is strictly prohibited. If you have received this email in error, > please immediately notify the sender via telephone or return mail. > > > > -- > Maarten Mennes, Ph.D. > Senior Researcher > Donders Institute for Brain, Cognition and Behaviour > Radboud University Nijmegen > Nijmegen > The Netherlands > > Google Scholar Author Link > <http://scholar.google.com/citations?user=pLlSTVgAAAAJ&hl=en> > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Head of Analysis, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [email protected] http://www.fmrib.ox.ac.uk/~steve <http://www.fmrib.ox.ac.uk/~steve> --------------------------------------------------------------------------- Stop the cultural destruction of Tibet <http://smithinks.net/> _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
