Dear SPM folks,
Is there an option in SPM (or a toolbox, or some other plug-in) for estimating the GLM using regularised regression? It doesn't need to be anything fancy, simple ridge regression would be fine. I appreciate that writing code to do this would be quite doable, but if there happens to be a pre-existing and battle-tested add-on already out there, then that would be better.
I've googled around and searched on this list and, as far as I can tell:
1. This isn't already built-in to vanilla SPM (or if it is, I am missing that fact).
2. Regularisation *is* already built-in for spatial normalisation, but that's not what I'm asking about. I'm asking about using regularised regression to estimate the GLM beta coefficients.
3. The list does contain some mentions of "spatial regularisation" in the regression, but I think they are just referring to the fact that spatial smoothing will tend to make the betas of neigbouring voxels become more similar to each other. That's not what I'm asking about either. I'm talking about a regularisation term lambda being used for the regression estimation of each individual voxel's beta value.
4. There exist pattern-recognition toolboxes, e.g. PRoNTO, which include regularised regression and which use SPM functions. However, as far as I'm aware this toolbox does not actually enable SPM itself to use regularised regression for calculating its GLM fits. Instead, it requires a whole separate analysis stream (I think). What would be nice, if it exists, would be simply to take an existing SPM design-matrix as it is, and estimate its GLM using regularisation.
5. I've also tried searching using the "-ization" spelling with a "z" instead of an "s".
If there happens to exist a regularisation plug-in which uses crossvalidation or some other method to find a good value of lambda individually for each voxel, then that would be even better. But that's asking a lot. My main question is simply whether any sort of SPM GLM regularised regression plug-in exists.
Many thanks,
Raj
--------
Rajeev Raizada, Ph.D.
Assistant Professor
Dept. of Brain & Cognitive Sciences
University of Rochester
Meliora Hall
Rochester, NY 14627
[log in to unmask]
http://raizadalab.org
|