[FieldTrip] Merging runs with different grad.tra matrix

[Konstantinos Tsilimparis]

Hi Jan-Mathijs,

Then would you suggest performing source reconstruction separately for each run and then averaging them in source space?

Best,
Konstantinos


________________________________
From: fieldtrip <fieldtrip-bounces at science.ru.nl> on behalf of Schoffelen, J.M. (Jan Mathijs) via fieldtrip <fieldtrip at science.ru.nl>
Sent: Thursday, November 7, 2024 8:55:54 AM
To: FieldTrip discussion list <fieldtrip at science.ru.nl>
Cc: Schoffelen, J.M. (Jan Mathijs) <janmathijs.schoffelen at donders.ru.nl>
Subject: Re: [FieldTrip] Merging runs with different grad.tra matrix

Dear Marisa,

You can use the option cfg.updatesens = ’no’ in the input cfg to ft_componentanalysis, which results in the grad.tra not being updated.

Yet, I would indeed be cautious with respect to concatenating across runs with a per-run ICA applied, even if you can bypass the error that is causing you problems. Specifically - although much depends on the downstream analysis you plan to do - differences in the unaccounted for noise subspaces may cause numerical instabilities, sensitivity reduction etc. (e.g. for data covariance inversion for beamformers, MVPA approaches etc).

Yet, in your case, since you also already applied a Maxfilter (per run) with ‘head repositioning’ (and perhaps even tSSS) any potentially unrecoverable woeful things may have already happened to your data prior to doing anything in FieldTrip. Strictly speaking, the Maxfilter per run also leads to a run-specific re-balancing of the channels, which  - depending on whom you’d ask - also requires an update to the tra-matrix, and for the tSSS would require an updated tra-matrix per time slice: however, if the Maxfilter has been applied in the Elekta software, the spatial (/spatiotemporal) projection information is typically not available anymore. On the other hand, if you put your trust in the Maxfilter, and you believe that the repositioning has stabilised the head position across the runs (and thus may have stabilised the relative position of any (neural or artifactual) source w.r.t. the sensors), you could consider doing a single run of ICA across the concatenated data. Also, you may anyhow consider to look here: https://www.fieldtriptoolbox.org/workshop/meg-uk-2015/fieldtrip-beamformer-demo/#part-3---reconstruct-single-trial-cortical-responses, which describes a PCA-based stabilisation of multi-run Maxfiltered data.

Best wishes,
Jan-Mathijs




On 6 Nov 2024, at 18:40, Marisa Monika Amalie Birk via fieldtrip <fieldtrip at science.ru.nl> wrote:


Thank you again for your time!

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20241107/6dac81cc/attachment.htm>