PLS for multimodal imaging
Posted on 12/06/09 07:31:20
Number of posts:
Dear Folks,
I am wondering whether it would be possible to apply your PLS package somehow to multimodal images? I have noticed that one can apply it to different modalities seprately but what about making a big matrix containing two modalities at a same time and apply SVD?
Thanks
Hamed
Untitled Post
rmcintosh
Posted on 12/07/09 12:47:07
Number of posts: 394
By different modalities, do you mean something like fMRI and ERP? If so, its possible to do so, but its not part of our GUI. The trick to merging datasets is to have one common dimension that can relate them. The easy one is subjects, but one can also do so spatially. For example, if you have fMRI and EEG, you can project the EEG solution into voxel space using some sort of minimum norm solution. This was an approach used by Valdes-Sosa's group in Cuba with multiway PLS.:
Martinez-Montes, E., et al., Concurrent EEG/fMRI analysis by multiway Partial Least Squares. Neuroimage, 2004. 22(3): p. 1023-34.
So, to answer your question, it is certainly do-able, but requires a bit of preprocessing to get a common reference for the data sets.
Randy
Untitled Post
as22kk
Posted on 12/08/09 05:11:18
Number of posts: 60
The recent paper (2009) published in NI would probably elucidate the application of partial least square for functional and structural brain images:
Linking functional and structural brain images with Multivariate network analyses: a novel application of the partial least sqaure method
(K.w.Che, et ....)
As Randy pinpointed, an easy way of dealing with multimodal images is to specify subjects as a common dimension between modalities.One option to go for this (if you are interested in a GUI package), would be to use Joint or parallel ICA ( http://icatb.sourceforge.net) which accommodates the need for analyzing multiple modalities. For example, parallel ICA in that package aims to find hidden factors from both modalities and connections between them.
Best
/Alireza