Back to the main page.

Bug 3434 - Virtual Channel reconstruction with sloreta

Reported 2018-06-20 18:14:00 +0200
Modified 2018-11-12 10:52:56 +0100
Product: FieldTrip
Component: inverse
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P5 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

- 2018-06-20 18:14:51 +0200

Created attachment 865 time courses: sensor, sloreta virtual channels and lcmv virtual channels Hi everyone, we've run into a problem using sloreta as a method to calculate virtual channel time series. Time series calculated by multiplying the sloreta-filters with the data result in weirdly noisy time-courses (as compared to i.e. lcmv filters, see attachment). The filters itself seem to be the problem - they behave very strange showing very different weights for neighbouring voxels, when sloreta should produce a smooth solution. Either we've completely screwed up the steps up to the forward model or something is going wrong within the calculation of the filters. We'd appreciate it so much if you could look into this! an example data-trial with headmodel, leadfield and gradiometer definitions as well as a test-script can be found here: Best, Lea

Jan-Mathijs Schoffelen - 2018-09-03 14:12:42 +0200

I don't know about sloreta, and in particular I don't know about the implementation in FieldTrip, which was contributed by Sarang Dalal, but it seems that you need to use some regularization in order for the solution to make sense, in cfg.lambda. Note, that 'eloreta' seems to give numerically sensible results, but this does not allow for a 'fixedori'.

Jan-Mathijs Schoffelen - 2018-11-12 10:52:56 +0100

closing time.