That is most likely unstructured gaussian noise. The best way to deal with that is with parcellation (averaging the timeseries within an area).
Peace, Matt. From: Romuald Janik <romuald.ja...@gmail.com<mailto:romuald.ja...@gmail.com>> Date: Wednesday, October 18, 2017 at 5:12 AM To: Matt Glasser <glass...@wustl.edu<mailto:glass...@wustl.edu>> Cc: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: Re: [HCP-Users] Noise in rfMRI I attach a picture with the timecourse of one grayordinate (demeaned and normalized to its standard deviation). The spikes that I wrote about are marked in red. They start below the mean and in one timestep go above 1 sigma and then again return below mean (i.e. below zero here). There are also other spikes in the negative direction there (unmarked) - like the huge one around timestep 18. This huge variability at the scale of a single timestep seems very much inconsistent with smoothing by HRF. Hence my question whether this is noise.. Many thanks, Romuald On Wed, Oct 18, 2017 at 12:52 AM, Glasser, Matthew <glass...@wustl.edu<mailto:glass...@wustl.edu>> wrote: Maybe if you post a picture I will know better what you were describing, as what I said below is general. Peace, Matt. From: Romuald Janik <romuald.ja...@gmail.com<mailto:romuald.ja...@gmail.com>> Date: Tuesday, October 17, 2017 at 11:23 AM To: Matt Glasser <glass...@wustl.edu<mailto:glass...@wustl.edu>> Cc: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: Re: [HCP-Users] Noise in rfMRI Dear Matt, Thanks for the references and answers. Given your experience would you consider the short spike that I mentioned as noise or this is not so clear cut? Many thanks, Romuald On Mon, Oct 16, 2017 at 8:54 PM, Glasser, Matthew <glass...@wustl.edu<mailto:glass...@wustl.edu>> wrote: 1. ICA+FIX removes the spatially specific structured noise in the HCP data, which includes cardiac, motion, and some kinds of respiratory effects 2. Ideally task fMRI data will be processed with ICA+FIX in the future (such data exist for a subset of the subjects on our local server, but it has not been run in ConnectomeDB yet). ICA+FIX clearly removes false positives and enhances statistical sensitivity of task fMRI data 3. It’s hard to determine a particular threshold where one should do lowpass filtering and be sure on is not removing some neural signal. Additionally, unstructured noise can be removed in other ways. Have a look at these references: https://www.nature.com/neuro/journal/v19/n9/full/nn.4361.html https://www.ncbi.nlm.nih.gov/pubmed/21854968 There certainly is random thermal noise in the data, but that can be reduced by parcellation or the Wishart Roll Off technique mentioned in the first reference. Additionally there is the issue of global noise and global signal in resting state fMRI data, which is the subject of a recent bioRxiv paper: https://www.biorxiv.org/content/early/2017/09/27/193862 Peace, Matt. From: <hcp-users-boun...@humanconnectome.org<mailto:hcp-users-boun...@humanconnectome.org>> on behalf of Romuald Janik <romuald.ja...@gmail.com<mailto:romuald.ja...@gmail.com>> Date: Monday, October 16, 2017 at 12:15 PM To: "hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>" <hcp-users@humanconnectome.org<mailto:hcp-users@humanconnectome.org>> Subject: [HCP-Users] Noise in rfMRI Hi, I wanted to ask a couple of questions on the sources of noise which can still interfere with true neural signal in rfMRI HCP data. My main question is in fact 3) below 1) Are the physiological (cardiac and respiratory) sources regressed out? It was not completely clear to me in the Resting-state fMRI at HCP paper of Smith et.al<http://et.al>., as this was written as optional (on page 160) - does the ICA-FIX remove it nevertheless? In fact when I looked at the frequency spectrum of the average signal over the cortex I found a noticeable peak around 0.29 Hz. 2) I understood that the task tfMRI data are differently preproccessed - there is no ICA-FIX and so the physiological signals are even more so there? (and the pattern of noise may be distinct from the rfMRI cleaned data? 3) In the resting-state paper, in the section justifying no low-pass filtering there is mention of thermal noise which becomes more important at higher frequencies. I looked at resting state data and for each grayordinate I subtracted its temporal mean and divided by standard deviation. Then I looked at local maxima (peaks) situated above 1 sigma. To my surprise I found that around 25% of those peaks are spikes of very short duration where the signal immediately preceding the peak and immediately following the peak is below zero. Are such spikes nonneural? Are these spikes examples of thermal noise? Such signals would be very difficult to reconcile with the smoothing induced by HRF (even the shorter version suggested for resting-state in the paper of Chen and Glover [BOLD fractional contribution...]? Is this correct? Many thanks, Romuald _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org<mailto:HCP-Users@humanconnectome.org> http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
