Back to the main page.

Bug 2007 - Proposed changes to ft_inverse_beamformer_dics

Reported 2013-02-23 18:12:00 +0100
Modified 2020-07-02 09:00:25 +0200
Product: FieldTrip
Component: inverse
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P3 enhancement
Assigned to:
Depends on:
See also:

Vladimir Litvak - 2013-02-23 18:12:54 +0100

Created attachment 425 Modified code I wrote a new DICS implementation for the SPM12 beamforming toolbox and I wanted to use the new ft_inverse_beamformer_dics. I needed to make some changes to the code (see attachment). Perhaps it's worth thinking about how these can be incorporated into this and other new inverse functions as the reasons for these changes are general. For now I'll use my modified version of the function. 1) I needed to separate filter computation from using the filters to beam. This is due to how the beamforming toolbox is structured but also has objective reasons also in FT (sometimes the data used for filter computation and the beamed data are different). Therefore, I added the option 'filteronly' to only compute filters and 'filterinput' to input filters instead of leadfields. 2) I wanted to loop over dipoles in my code (e.g. to use SPM GUI progress bar). Inverting Cf and checking its rank in every call makes this very slow. Thus I had to add an option to provide invCf as an input. This is also useful because there will be cleverer ways of inverting Cf in the toolbox and we'd like to be able to use them instead of the default. Vladimir

Jan-Mathijs Schoffelen - 2020-07-02 09:00:25 +0200

Revisiting some open bugs that have to do with inverse modelling. There has been substantial recent work (spring/summer 2020) that has improved the code base for inverse modelling, making the flow of the code much more logical, and the different implemented inverse algorithms much more consistent relative to one another. In addition, earlier work (spring 2019) has already made the matrix inversion scheme much more flexible. I would assume that this report here can be closed, since it's probably not relevant anymore, or at least it's become a bit stale. Feel free to open a new issue about this on github if needed.