Functional Neuroimaging - Science topic

Maybe this paper will help here: https://pubmed.ncbi.nlm.nih.gov/17258472/

cheers

Michael

Assuming the mechanism is "missing at random" (MAR) or "missing completely at random" (MCAR), you could use full information maximum likelihood (FIML) or multiple imputation to address missing scores. Both methods are based on the same mathematical theory and often lead to similar results.

Enders, C. K. (2010). Applied missing data analysis. Guilford Press.

Hello Amin,

just wonder if the empty array is still working for you?

With SPM 12 v. 7487 I get a message:

Item 'Onsets', field 'val': Size mismatch (required [Inf 1], present [0 0])

This suggests that at least in the somewhat recent SPM versions there is an internal check for the size of the onsets array.

For example, you could invent a measure based on cluster coefficients.

You could consider all vertices that lie in a shortest edge distance up to a certain maximum, and count all edge in it.

Also, it can be interesting to determine the number of cliques a vertex takes part in. Or the distribution of the number of vertices in these cliques. Or the overlapping percentages of any pair of cliques a vertex is in.

Regards,

Joachim

Will message you.

CT brain images database:

Upon more thinking and reading, I think the easiest solution is to use fslmeants to simply extract the mean beta weight across all voxels from the Stage-2 Output Maps. I did multiple sanity checks and find that the mean values and individual beta maps are fairly comparable (see bottom row of attached screenshot).

It also helps to apply smoothing prior to group ica (kernel = 3mm) to get cleaner components that more resemble functional networks in case anyone runs into this problem in the future.

There are many books out there on the control of emotion. Work I did years ago delineated how the training of combat soldiers had such an effect.

I try to answer not with what i studied for my neurology exam.

Reading is a really complex function, which, in the normal person, actually stresses out several different brain skills as auditory, visuospatial and logic ones.

Basically in the individual without hearing problems, reading, both with loud voice or in his mind, means transforming the graphems into their correspondant sound, and after having "heard" it understanding through wernicke's area ad insular language centre.

People who, themselves, have hearing problems since their birth, are probably able to skip this two-phase process and, directly send the graphems to be logic matched with a meaning in the language area, without converting them into sound.

Sounds clear that the temporoparietal networks where the recognition of graphems as meaningful visual elements must function in order to keep this process, and it is here, where the visual part of reading happens, that dyslexia happens, if something goes wrong.

You can find these information on Adams and Victor's principles of Neurology.

Good luck for your thesis!

Hi, EEG is already used for dynamic connectivity analysis and in my opinion it is much better tool for this (because of temporal resolution). The neuronal connectivity in EEG is also known for much longer than fMRI (however named differently, as coherence or DTF). There are few things that you have keep in mind to do this correctly:

1. You need high density EEG - at minimum 32 channels, recommended 64 and above

2. Very good data quality (i.e. high SNR).

3. There are a lot of papers suggesting that EEG connectivity analysis should be performed using source space and not sensor space. So first data has to be recomputed to source space.

I

think fieldtrip has some solution for EEG connectivity analysis if you need freeware tool. You might also check BESA Connectivity

PM me in case you need any more info about Besa Connectivity (I do not want to make advertisements here :)

Dear Edgar Guevara , the fMRI data is not resting state, but rather from a visual motion perception task. I forgot to mention that in the text.

Hi,

as said and expected the underlying functions devoted to the .fif reading should be the same in both toolboxes.

 If you want to dig deeper, take a look on how the example file 

mne_ex_read_raw

uses

fiff_setup_read_raw

So, if i do this now on a random fif file on my computer, I get

info = fiff_setup_read_raw(fileName)

the output  is:

Opening raw data file blahblah.fif...

Read a total of 13 projection items:

generated with autossp-1.0.1 (1 x 306) idle

[13 times repeated] ...

Range : 16000 ... 654999 = 16.000 ... 654.999 secs

Ready.

info =

fid: -1

info: [1x1 struct]

first_samp: 16000

last_samp: 654999

cals: [1x467 double]

rawdir: [1x639 struct]

proj: []

comp: []

As you can see, the first sample in my measurement is 16000 and the last is 654999, but I'm neither surprised nor worried of this behaviour: this happens just because, in this particular acquisition file, between the "Go" button press and the "Record raw" ticking passed exactly 16 seconds. In this 16 seconds very likely someone just checked the HPI, and scrolled through the MEG channels to take a look, and then pressed "Record raw". These same numbers you will get as well if you use elekta's very own "show_fiff" utility on the acquisition console. This is a "finnish peculiarity", dunno otherwise how to call the fact that in the final file they start counting from 16000 instead of 0 ... 

So, my take is that you had 10 seconds (at 1 kHz sampling) before ticking the "Record raw" check box. *But* if you say that you have missing conditions/trials/triggers in your final list, then the only answer that I have is that someone started the stimulation paradigm *before* ticking the "record raw": once in a while happens also to me with young students . Don't put the blame on the analysis software, or on the acquisition, go back in the lab and slap someone randomly ... :-)

HTH

Hi Amir,

Concerning the BMS, you could have a look at the VBA toolbox (http://mbb-team.github.io/VBA-toolbox/wiki/BMS-for-group-studies/). It is independent of SPM, so you will have to extract the model evidence of each estimated model-subject (DCM.F if I remember correctly), but you can script all types of model comparison (RFX, between condition, between groups).

Also, as a general rule, and even if all subjects are best described by the same model, always work with the Bayesian Model Average (BMA) for your post-hoc analyses on the parameters.

I hope this helps.

Best

LIonel

As far as I know when you use ROI analysis, you have one averaged value per ROI (like ROI is you one big voxel, you assume that all voxels in this ROI do the same work (function) and when you perform SVC, you still have all your voxels with their values in you volume if interest but after you chose this volume for analysis you have fewer voxels than in the whole brain and you activation within this volume of interest is able to survive after correction for multiple comparisons. I think - this is the main difference.

You'd still need to figure out how to hold the head plate in place (assuming you didn't want to buy Neurotar's system). 

I am not sure that I understand the question correctly, but you can always write recursive code in a function and then simply call the function. Otherwise you can build a whole class around the progress bar and update the class whenever needed.

If that does not answer your question, a little more detail to your problem would be beneficial.

Cheers,

Martin

Hi Rhiannon,

in my experience, MEG and EEG are friendly techniques in case of pregnacy because they both only sense the electromagnetical activity emerging from the brain. As Stephen mentioned MEG have been perfomed already in fetal studies. Personally I worked with Magnetocardiography in pregnet women to record the cardiac activity of multiple fetus (twins and triplets) during weeks without secondary efects (Processing the magnetocardiographic signal in the identification of fetal and maternal heart beats in a triplet pregnancy).

However, in case of CT, fMRI, PET and NIRS I have my doubts. 

Klein and  Hsu wrote an interesting article touching this topic. This can be a useful reference for you. Take a look: Neuroimaging during pregnancy (http://www.ncbi.nlm.nih.gov/pubmed/22113508).

Here you have another reference about fetal MRI: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4515352/ .

There are studies and guidelines already for some neuroimaging techniques applied during pregnacy:

Compendium of National Guidelines for Imaging the Pregnant Patient:  http://www.ajronline.org/doi/abs/10.2214/AJR.10.6351

Imaging of Pregnant and Lactating Patients: Part 2, Evidence-Based Review and Recommendations: http://www.ajronline.org/doi/abs/10.2214/AJR.11.8223

On the other hand, here is an rticle about side effects:

I hope these information can be useful to you. Probably you already read those articles.

Have a nice day. All the best for your research !

To me, generating data from different methods and checking the consistency of the results is a good way to validate the robustness of your method. Maybe our fields are different, in our field (neuroscience), robustness is not enough, people would like to see that you use some results to show your finding is not from noise. To do so, you may need to test the biological meaning by direct (i.e., by changing the subject's status to see if status changing modulate your result) or indirect ways (i.e., correlation with previously widely-used metrics).  These are related to "validity".   Hope it helps. 

Dear Xue,

If you don't have a fieldmap, you might try to use the SPM12 normalization. This works best when your images are more or less spatially aligned.

-Segment the T1-weighted image (with SEGMENT in SPM12)

-Combine the generated tissue maps to a 4D niftii-file (e.g. with fslmerge)

-Use the combined images as tissue probability maps in the SPM Normalization, 

-As data best select the mean functional image

-When your images are more or less aligned, select 'no affine registration' under 'affine regularisation' .

-Run and hope for the best

Kind regards,

Mathijs Raemaekers

I guess I found the reason for the discrepancy between the ALE output file and the Yale web application. In the output file, the nearest gray matter is depicted, while the web application shows me the single point. When testing for both with the Talairach daemon, I get Insula (when searching for the nearest gray matter) and inferior parietal lobule (when searching for the single point) using the same coordinates. 

False sentence is totally wrong i.e. He was arrested for giving false information on his application for a passport. Incongruous mean unsuitable, inappropriate, inconsistent, inharmonious, incompatible, conflicting, discordant. For example: It seems incongruous to have an out-of-shape and overweight editor of a fitness magazine.

Hi Jason,

Have you considered a mental arithmetic task? The difficulty of serial subtractions can be varied by adjusting the length of numbers in the subtrahend and/or minuend. Difficulty might also be manipulated by adjusting the the time allowed to provide an answer, but this could introduce unwanted emotional activation.

Best,

Mal

Sorry i'm not familiar with those instruments.  Most companies can answer simple questions by phone or e-mail.  Lacking that, a sales representative may be willing to stop by and give some advice.

Robert M. Reinking

Maybe these papers can bring some clarity into the matter:

Yes, this is 4000mV at detector (due to light intensity) that goes beyond the Analog to Digital (A/D) converter limit and will results in saturation for higher values. You need to adjust LED current and brightness to lower the light intensity at the detector.

in my opinion the best way is to have enough trials in order to have the possibility to compute statistical analyses only on the remaining artefact-free trials....

Hi

If you want to model the contrast (congruent +incongruent)>baseline using a Tcontrast you could use either the vector 1 1 -2 or 0.5 0.5 -1 and if you want to model the opposite (i.e. baseline>(congruent+incongruent)) you have to change all signs in the expression (i.e. -1 -1 2 or -0.5 -0.5 1). When you see contrasts as 1 1 0 or something like that, they could be F contrasts that are a little bit different and are used to test a hypothesis about general effects, independent of the direction of the contrast!

Hope this helps!

hi there Muhammad,

Depends what you want to do. I've used both SPM and AFNI a lot for fMRI. In my opinion AFNI is more versatile in terms of customizing an analysis. It has many more tools and functions for visualizing your data at an astoundingly detailed level. I believe that it's stronger with resting state analyses (see ANATICOR, i.e. Hang Joon Jo, 2010) and with nonlinear registration (i.e. 3dQwarp) and can do many advanced kinds of regressions (one may need 'R' installed, however). Although afni_proc.py has streamlined and standardized many kinds of analyses in AFNI, it perhaps requires more a little more tcsh scripting than your average SPM user uses Matlab scripting. SPM has some analyses available that no other package has (like dynamic causal modeling) and yes it's true there are many extensions that the SPM community has contributed ( http://www.fil.ion.ucl.ac.uk/spm/ext/ ). SPM's also stronger with modalities other than fMRI. However, it doesn't integrate as fluidly with large parallel computing clusters as AFNI does. Also the packages have slightly different 'ideological foci' in that SPM's authors and community is a little heavier on computational modeling than AFNI's, which in turn is a little more focused on very carefully looking at and understanding one's data. Because SPM's community is larger there are a few more workshops available around the world to learn it. Both have active listservs. After using SPM throughout my doctorate, I was quite pleased with AFNI once I got a hang of it during my postdoc. I plan to return to SPM for certain things, however. Like I said, depends on what you want to do.

-Salvatore (Sam)

Dear  Miroslav:

 I will give an indirect answer to your question. In Remote Sensing, we always use the stacked images in the form of multi-spectral and hyper-spectral images and there is a plethora of literatures that can be adapted to your field of study. Neuron or Neural networks are computational models to capture the underlying dependency in any data set in the form of supervised or unsupervised modes of learning. Hopefully this answer may shed some light to your question.

Regards,

Gamal

Very good question - I'd say it depends on your definition of plasticity and the response you expect. A few thoughts -

If you expect additional regions to be recruited or existing regions to be more strongly recruited, I agree suprathreshold counts and some other measures are problematic, one approach we've tried is SPM's multivariate Bayesian decoding:

Connectivity approaches approaches can also be useful, see our recent discussion of the related issues of compensation and reorganisation:

Lovden and others (2010) also have provided some definitions of plasticity in relation to training interventions and the kinds of neural responses to be expected:

I would recommend using the CONN functional connectivity toolbox, which allows you to view and export connectivity results with ease, and which converts r-values to normalized z-scores.

As David mentioned, what do you mean with intra-psychic conflict? For example the dorsal anterior cingulate cortex (dACC) has been shown to be associated with conflict monitoring.

If you are willing to get into a bit of programming, you could try the vmtk libraries (http://www.vmtk.org/). The authors would probably mix in 3d Slicer for free registration and visualization. They seem now to have a commercial offering which presumably makes life simpler, but not free. 

The PAR/REC format is proprietary to Philips, however, I know that this is also possible to tell the system to output nii.  

Thank you Marco and Wanyong for your answers !

Virtual Fly Brain were certainly setting up to do something like that a couple of years ago. I've not looked into it too much since, but you can try it out with the link below

There is very little that the number of components found could tell you about brain functioning or connectivity, since many of the components could be related to noise, artefacts or be vascular components. You should take some considerations into account. Did you perform group ICA with temporal concatenation? Did you use automatic estimation of the number of ICs?

If you really want to compare the integrity of the resting state networks in a group of patients vs. controls your approach is not adequate.  I suggest that you go for a entire sample group analysis, and after identification of the real resting state networks you apply a two-groups differences analysis. Take into account structural differences into account. In the end, differences found would reflect differences in network integrity/pattern. 

It makes no sense to use a "small volume" that includes parts of the image where brain activation cannot possibly occur--i.e. in white matter, bone or CSF. Thus the spherical "small volume" normally used in SPM is fundamentally inappropriate, since it would need to be less than 3 mm diameter to not include WM, bone or CSF. It is greatly preferable to use a shaped small volume that includes only grey matter, from one specified and well-defined cortical area, or one deep nucleus. This can only be achieved when a suitable individual-brain cortical and deep brain parcellation has been performed. See my recent paper 'Comparing like with like: the power of knowing where you are' (Turner R, Geyer S. Brain Connect. 2014 Sep;4(7):547-57) to see further details of this approach, which takes advantage of the unique sensitivitiy of MRI to details of myeloarchitecture and iron distribution. To use such small volumes considerably increases the experimental power, and enables well-grounded averaging across subjects and comparison between subject groups. Once this approach has been widely adopted, we will be able to place far more confidence in fMRI findings, in any type of study

Hello

Dear friend, 

See This links. I hope be useful for you.

Best Regard

You may be interested in these 4 entry points. Contacting these authors should narrow your search.  Be also aware of the variability of the vascular territories.

I am also looking at some similar material.

Ref1.  probabilistic map produced by Michel Dojat in Grenoble france (GIN, Grenoble Institut des Neurosciences INSERM : U836). The digital Atlas of the Blood supply territories of the brain (BST), derived from the 12 printed serial sections in the axial plan developed by Tatu et al. The atlas fits the Talairach space, this 3D atlas is used to determine the stroke subtype

see : Yacine Kabir, Michel Dojat, et al. Multimodal MRI segmentation of ischemic stroke lesions. Conf Proc IEEE Eng Med Biol Soc. 2007 ; 2007: 1595–1598

Ref. 2: the digital probabilistic map of PCA infarcts produced by Thanh G. Phan et al (Digital Map of Posterior Cerebral Artery Infarcts Associated With Posterior Cerebral Artery Trunk and Branch Occlusion - Stroke 2007;38;1805-1811). They register their PCA lesions with the MNI template. It is not a vascular territory map but a map of probability of stroke.

Ref. 3: Another probabilistic map for ICA produced by Jae Sung Lee et al in Seoul.: Probabilistic map of blood flow distribution in the brain from the internal carotid artery. NeuroImage 23 (2004) 1422– 1431.

Ref. 4: Analysis of ischemic stroke MR images by means of brain atlases of anatomy and blood supply territories. W Nowinski et al. Academic Radiology. 13, 8, Pages 1025–1034, August 2006. Their approach is a atlas-to-scan transformation (mapping), quite the opposite of what you may want to do (mapping each scan to the registered atlas).

This is the review article that contains information and references of importance. Note that medial and lateral striatum are different in terms of anatomy, physiology, and behavioral function.

Hi Carmen,

faces are always problematic due particularities of visual perception. I think you may find interesting these references:

Little, A., DeBruine, L., & Jones, B. (2005). Sex-contingent face after-effects suggest distinct neural populations code male and female faces. Proceedings of the Royal Society B: Biological Sciences. 272, 2283 – 2287.

Loffler, G., Yourganov, G., Wilkinson, F., & Wilson, H. R. (2005). fMRI evidence for the neural

representation of faces. Nature Neuroscience, 8(10), 1386 –1390.

Webster, M. A., & MacLeod, D. A. (2011). Visual adaptation and face perception. Philosophical Transactions of the Royal Society B, 366, 1702–1725.

Carbon, C.C., Grüter, T., Weber, J., Lueschow, A. (2007). Faces as objects of non-expertise: Processing of thatcherised faces in congenital prosopagnosia. Perception, 36, 1635 – 1645.

Gauthier, I., & Tarr, M. J. (1997). Becoming a `Greeble' expert: Exploring mechanisms for face recognition. Vision Research, 37, 1673 – 1682.

Thompson, P. (1980). Margaret Thatcher: A new illusion. Perception, 9, 483 – 484.

Webster, M. A., Kaping, D., Mizokami, Y., & Duhamel, P. (2004). Adaptation to natural facial categories. Nature, 428, 557 – 561.

If you can't find any of them, please don't hesitate to contact me.

Most likely, in your case, when you run a general linear model (GLM) analysis, you are attempting to predict the voxel signal using a vector of onsets convolved with some hemodynamic response function. It's fundamentally the same as when you are carrying out any other linear regression: You are predicting some variable (Y) as a function of a predictor variable (X): Y = B(X)+E -- Y is a function of X, plus some variance, (E). B is simply the coeffcient of the predictor variable that leads to the best estimate of Y. In the case of fMRI, X is the idealized time course for the onsets of a particular condition. When B is high for a voxel, that means that the activation for that voxel closely follows the idealized time course for that condition. If B is near 0, that means there is little to no relationship between voxel activity and the idealized time course. B can also be negative, meaning that it is inversely related (e.g, whenever that particular condition occurs, activation actually decreases).

I think the co-activation network across participants might be the most simple and direct approach. This method seems to apply to any fMRI design.

Yes, fMRI analyse time series data. The conventional approach is to apply general linear modelling to all voxel time series and check which voxel(s) have a significant correlation with the contrast-of-interest. This approach is univariate in nature, i.e. the analysis is performed on voxel-by-voxel basis.

However, there is a new trend to consider signals from multiple voxels as a pattern and feed them into machine learning algorithms. Using machine learning algorithms on fMRI BOLD or EEG/EMG signals is a very new and active field in the imaging community. People had successfully made classification for clinical diagnosis, predict active brain states using activation paradigm (brain decoding), and even develop brain-machine interface with real-time fMRI.

Just a few seminal papers you probably would interested in:

Soon, C.S., Brass, M., Heinze, H.J.& Haynes, J.D. (2008). Unconscious determinants of free decisions in the human brain. Nature Neuroscience 11, 543-5.

Reconstructing Visual Experiences From Brain Activity Evoked by Natural Movies

Shinji Nishimoto, An T. Vu, Thomas Naselaris, Yuval Benjamini, Bin Yu & Jack L. Gallant (Current Biology 2011)

I have an MNI one. Feel free to email me and I can send it.

I am assuming you're looking to lateralize language for presurgical planning? One simple test, that has been fairly successful includes flashing a description of an object for the patient to read. His/her response will be to silently name the object to themselves while in the MRI. The control portion of the test could be nearly anything, but usually staring at a cross on the screen of their goggles works well. A good description of the task can be found in "Language dominance in partial epilepsy patients identified with an fMRI reading task" Neurology (2002) 59, 256-265. Hope this helps!

Dear Michael,

This great paper and its references should give you a good start into magnetic resonance spectroscopy:

Rae, Caroline D. „A Guide to the Metabolic Pathways and Function of Metabolites Observed in Human Brain (1)h Magnetic Resonance Spectra“. Neurochemical Research 39, Nr. 1 (Januar 2014): 1–36. doi:10.1007/s11064-013-1199-5.

Free Software exists, e.g. http://tarquin.sourceforge.net/

Many groups use LCmodel since it gives a high reprodicibility (unfortunately the price is quite high, too: http://s-provencher.com/pages/lcmodel.shtml)

All the best

Steffen

A quick tool for looking up anatomical brain locations using MNI coordinates is MRIcron (http://www.nitrc.org/projects/mricron), a free and highly useful tool. Within the program, you can open one of the pre-installed templates such as the AAL (automated anatomical labeling atlas) and use the pull down "view" menu to enter your MNI coordinates. It will show you the location and the anatomical name based on the the atlas. It comes with a Brodmann atlas as well.

Please contact Prof. Michael Deppe in Münster, Germany via

deppe@uni-muenster.de or Michael.Deppe@ukmuenster.de

PET scan is not the final answer in Alzheimers disease. The information that we get from the history as how the disease started cannot be provided by any investigational modality. Moreover even after PET scan we may not be able to differentiate between frontal variant Alzheimers disease and behavioural variant FTD. Even in those cases of FTD which satisfied the diagnostic criteria with supportive imaging and PET findings, a significant number of cases turn out to be alzheimers disease finally on autopsy. So everything is not black and white here.

1. I do "Motion scrubbing" or "Motion censoring" in a non-destructive manner. By that, I mean that I never delete the high motion volumes but I ignore them in my regression analyses (nuisance regression, connectivity calculations, or activation calculation following your tasks...).

2. Depending on your TR, I would either ignore the one TR or the one TR plus one before.

3. For Task-fMRI, I have seen thresholds ranging from 1mm to 0.5mm while for resting state my choice is usually 0.25 or 0.2mm

4. I believe that the function I use for my fMRI analysis only ignores the time without removing them so no interpolation seems necessary. I believe this is true because we use scrubbing based on volumes and not voxels.

5. I think it is getting more and more common to exclude a subject if more than 50% of the volumes are censored and/or less than 5 minutes of data is remaining (based on resting state fMRI). Now if you have only have a 5 minute run, I would probably look at the distribution of how many volumes are censored for each subject and pick a threshold that keeps most of the subject (I don't think I would go lower than 75% of the run remaining).

Below is one of the better papers comparing covert and overt naming in the scanner. Relatively artifact free fMRI data can be obtained with overt naming especially when using an event-related design:

Procedures vary from site to site. You can learn about the procedures at our 7T MRI site here: http://www.imago7.eu/ENGricerca.html

In short, researchers that have interest in using the resources of Imago7 are invited to present their project to the Scientific Committee for preliminary feasibility assessment: segreteriaRM7T@imago7.eu

Then, fees will apply depending on the type and number of experiments.

Good luck!

Please provide operatin system and error messages.

and more..

Bayesian model selection for group studies

Klaas Enno Stephan, Will D. Penny, Jean Daunizeau, Rosalyn J. Moran, Karl J. Friston. NeuroImage 46 (2009) 1004–1017

[... Bayesian model selection (BMS) is a powerful method for determining the most likely among a set of

competing hypotheses about the mechanisms that generated observed data.]

enjoy :D

Alfredo

Dear Anoop,

The recommended protocol depends on the capabilities of your scanner (field strength and type of gradients) and on what you want to look at and how you want to analyze your data. Discuss this with your local physicist, if possible. Here is a set of settings that I think one can consider pretty standard. In a crossmodal audiovisual design you are probably interested in whole brain coverage (visual cortex, auditory cortex, but also frontal and parietal areas).

I would go for an EPI sequence with 3x3x3mm resolution, which you can achieve with 3mm slice thickness, and a field of view of 192mm (if you use a 64x64 image acquisition matrix, which is most likely among the standard sequences on your scanner). Put 10-15% gap between the slices.

Next check with a participant inside the scanner how many slices you will need to cover the entire brain. 32 slices should be alright. Keep the number of slices fixed for all participants in your study.

Next TE: depends on the field strength of your scanner, on a 3T 30ms is a good start, on 1.5T I would go a bit higher.

TR: Depends on the type of design. In an event-related design you may want to strive for a shorter TR (2000ms is a good value to start), but if you use a block design, you can afford longer TRs such as 2000ms to 3000ms (longer TR gives you more signal).

Flip angle (FA): Depends on the T1 of the tissue and on TR (with shorter TRs requiring smaller flip angles for optimal performance). A reasonable setting for TR=2000 is FA = 77 deg; for TR = 3000 FA = 85 deg.

This is just a start, there are many ways to optimize this.

Best,

Jens

Depends a bit on what you want to do and know. If you want to control for the effect of covariates on single subject data, then it's better to specify these in a 1st lvl design. If you want to know the effects of covariates in a group analysis, you would specify those in a 2nd lvl design. However, if there are large differences in RT between subjects, you may wonder, whether the mean RT says anything useful at all.

Cheers,

Wouter