Hi Aljoscha,

Thak you for your reply. I just forgot to mention that it is an fMRI study. I want to do a functional connectivity analysis usig CONN toolbox. It has diffrent option for analysis type that I haven't decided to choose which of them yet.

Many thanks,

Haleh

Your trials (or blocks) are too long as you have also mentioned (a bit tricky for working memory tasks which I would like to know more about the task and trials and why they are long), making it susceptible to increased preprocessing errors if noise correction and noise source reduction have been performed separately or nominally beforehand. Also, the main restraint in your experiment is the concern regarding head motion though you can remove it mostly here and with CONN.

The point is that other than having controls and rsfMRI cases to ensure the statistically reliable comparison, the use of toolboxes such as CONN for correlated and anticorrelated networks can ease this knotty preprocessing.

I think that it is doable, just suggest that you look for adjusting the temporal window for your BOLD signal contrast first. You may already now better than me about the first and second level analyses of random effects with your resting and task trial data. I am not sure though about whether these also depend on your ROI and/or voxel estimation. But you may want to try and use different features that similar papers have implemented such as graph analysis, vox-vox analyses and compare to various ROI sources if applicable.

Dear Bahman,

Thank you very much for your reply. Yes noise reduction will be harder for long inter trial intervals. Each trial consists of an encoding, a distraction and a retrieval part. It is an event related paradigm with fixed sequence of events whithin each trial. I also have different types of trials that are randomly distributed.Hence the time difference between similar events (e.g. retreival event) across different trials is very long (35.5 s or more). My main question is whether this long interval also makes predicting the shape of hrf harder for each event. I have used a jittering in a way that the length of each trial is a multiple of TR plus 500 ms.I was hoping that it gives me four points of hrf for the retreival event with the duration of 2s. Yet I'm not sure how effective this jittering really is. I was wondering whether I can predict the shape of hrf in each region and the correlation between hrfs of different regions efficiently with this paradigm?

Many thanks,

Haleh

Thanks for the details, Haleh.

Well as I expected, you are using ERF of randomly distributed trials with apparently convenient jittering of your ITIs that would naturally increase the statistical power (and possibly more statistical efficiency) of your post analysis, though blocked disign could have had less difficulties after all.

Convolving stimulus with HR could lead to the anticipated time-series of MR. Now just checking to be sure that you have done the following: I suppose that you have already used randomized SOA, and also applied high pass filter to minimize the loss of any signal.

Of course that long ISI would lead to more versatile HR shape, but I think that the jittering that you have applied would uncover HRFs that you wouldn't sample at all with fixed ISI (there is an exception that I mentioned at the end).

I think that you shouldn't be worry about the prediction of the shape of your HRFs as you might want the peaks higher to cause more statistical power. There is a certain trade off between statistical power and efficiency when you apply jittering. Probably it is even OK if you got different HRF shapes for each voxel. It is inevitable that sampling each point of the HRF fewer times by varying ISIs would increase the statistical efficiency by losing power, which does not matter!

(In the meantime, you may want to use AFNI's 3dDeconvolve -noinput option, which will take a given experimental paradigm and calculate how good it is in terms of power and efficiency.)

Yet you should be more sure about a reliable shape of every HRF as you sample as many points as possible. Your jittering as addition of 500 ms to the whole 2s event seems reasonable enough to produce a reliable window to minimize the loss of any possible elusive HRF peaks.

However, as I studied somewhere, if you're applying a new paradigm that you are not sure whether you may acquire any activation at all, you would probably better off applying or at least trying a block design and a fixed ISI to maximize your detection power.

I should add that as you know, suitable jittering depends on a variety of parameters such as the paradigm itself, what you'd like to control in it and the effect you want to see, etc. Check Liu's experimental design toolbox and study (linked below). You may find it useful too.

I read somewhere that you may get better temporal resolution compared to your TR by not sampling BOLD signal each time you present a stimulus! This will increase the chance that you get a good shaped HRF. But the downside to this is losing the statistical power...which as I mentioned probably does not matter compared to the gain that you will get!

1. Check the following for details about ERP fMRI design:

Article Efficiency, power, and entropy in event-related FMRI with mu...

Article Stochastic Designs in Event-Related fMRI

Article Optimal experimental design for event-related fMRI

2. Check this for better a view on jittering issues:

Article Temporal Jitter of the BOLD Signal Reveals a Reliable Initia...

Thank you for your inclusive reply Bahman. I have already collected the data and I have done a univariate GLM analysis. Based on our discussion about the shape of hrf I think I go a head and do some connectivity anaysis too.

Thanks for your time,

Haleh

Thanks, Haleh.

Although you may find further consultation helpful too, I think that you are all set as you have already done the analysis.