Unified Theory of FMRI?
Submitted by bpittman.
on 2005-12-20 14:01.
Don't we wish.
What are the fundamental principles, that most people could agree with, on which a theory of FMRI data analysis could be built? I can think of two (so far)
- Hemodynamic Response Function: FMRI data is too noisy, so we have to repeat many trials of each stimulus class, so we have to assume that the results are the "same" each time and then do some statistics. The "sameness" assumption, plus the assumption of additivity, is the HRF principle.
There are many potential ramifications of this principle for analysis, the simplest of which is the current favorite analysis for FMRI papers ("GLM"). Other possibilities include:
- Assuming the HRF is the same unknown function across the whole brain, with unknown amplitudes in each voxel; leads to a kind of principal components deconvolution.
- Assuming the HRF varies randomly a little between stimulus repetitions. Independently or coherently between voxels.
- Blobs: That is to say, we are looking for regions of activation, not randomly scattered junk. I'm not sure of the ramifications of this, other than that we need to come up with a coherent scheme for describing possible spatial activation patterns; then the job of choosing amongst them will be soundly based. Wavelets? On the cortical gray-matter shell/surface?
(i) a coherent set of models for FMRI data upon which
(ii) a coherent set of algorithms for extracting information can be based?