I don't think there can be an 'ideal' number as such, theoretically you should be able to localise the sources for single trials as well. Numerous factors can influence that, e.g. the MEG system you use (its dynamic range and SNR), different inverse solutions, trial length, individual anatomy etc

The easiest way to determine that would be to see the relevant literature for trial number and inverse solutions. 

Can you tell us more about your experiment? and the scanner used?

Hi Kanad, we're using the 306 channel Elekta Triux MEG. We're using a facial emotion processing task, currently working on 128 trials per condition (comprising 64 'targets' which are of interest). We're got 6 conditions and current experiment time is around is ~40mins. 

Hi Tamsyn, 

You can read Attal and Schwartz (2013) who have used minimum norm approaches to localise deeper sources and Luckhoo et al (2012, NeuroImage) who have used beamforming and temporal ICA. I think both the groups have used CTF systems (axial gradiometers) which provide a better SNR for deeper sources due to its base distance. So that is something worth considering. 

From what you've described, you should get something like a 3-3.5sec trial? That makes it ~180s worth of events of interest, which I think should provide you a good enough SNR.

Kanad 

Hi Kanad, 

Thanks for for the advice. Trials are 2.5s (1sec presentation time, 1.5sec ISI). Other literature (albeit limited) suggests trials of between 40-70 are ok. I just thought it would be good to get others opinions.

thanks for the help,

Tamsyn

Hi Tamsyn,

just to add to Kanad's advice: The Triux has planar gradiometers and magnetometers. The former indeed have a low(er) sensitivity for deeper sources, however the magnetometers may be even better suited than axial gradiometers - at least theoretically. In practice, SNR, method, etc. of course heavily influence the results.

Stefan 

Hi Tamsyn,

this is a bit late, but the "quality" of your source analysis depends highly on the method of choice. You mention rather long trials, but not whether you intend to use evoked activity (as used in dipole modelling, Minimum Norm or LCMV beamforming) or oscillatory activity of a specific frequency (as used in DICS beamforming or Sarang Dalal's frequency adaptive beamformer). As Kanad mentioned, different approaches have been shown to be able to localize deeper sources, but the better you know what you are looking for, the better your results will be.

From a more informal point of view, I'd say that 40 - 70 trials a fine for the localization of an ERF (it worked for me with an LCMV at least), but for oscillatory activity, I'd at least double the number of trials (again, this depends highly on the signal you are working with).

Good Luck

Dear Julian,

Thanks very much for your advice. We are interested in looking at gamma synchrony and have around 128 trials per condition of which we are looking at correct vs incorrect responses (there is a target to which the participant must label as being either the target or a non-target, thus 64 targets, 64 non-targets per condition). We are working with a group of psychiatric patients so cannot extend the experiment time much longer than what we have now. I'm hoping that 128 trials per condition will be sufficient for an analysis of gamma synchrony in the deeper regions.

If you have any more advice I would very much appreciate you sharing it (I am new to MEG).

Thanks and kind regards,

Tamsyn

Hi Tamsyn, 

Just like to add to whats been suggested: presence of artifacts can heavily influence deep source localisation so make sure that you think about a good data 'cleaning' strategy.

Given that you have sufficiently long trials and are interested in oscillatory activity more than evoked, DICS (or LCMV) would be your best bet. And they are neatly implemented in FieldTrip. Apart from your standard artifact removal, beamforming will suppress any additional noise which is always a plus. Although, depth weighted minimum norms are nice and easy, they are very sensitive to any additional noise and this will influence your localisation. 

Im sure your lab will have some standard procedure but if not then, irrespective of the inverse solution you use, make sure that you record additional ECG/EOG data, going beyond this a Cz or a Pz electrode even. You will be amazed to find out the number of ways with which you will be able to utilise this information.

Have you recorded a pilot yet? Do analyse that dataset in its entirely (and tell us what you find or if you get stuck) before you start running the experiment. 

Kanad 

Hi Tamsyn,

Kanad makes a very good point, the more info you record, the easier the artifact correction will be. Again, as Kanad mentioned, check out FieldTrip. In the FieldTrip tutorials, you'll find detailed instructions on how to perform the source analysis.

If you actually know what (gamma) and where (amygdala) you are looking for an effect, I suggest looking at NutMeg, which is a free Matlab toolbox heavily integrated with SPM and FieldTrip. Sarang Dalal's frequency-adaptive beamformer is implemented in this toolbox. Another approach would be to place a so-called "virtual electrode" into your volume of interest and project sensor-level data onto this source. I suggest checking the FieldTrip tutorials for the latter approach.

Good luck