Dear FSL users,
We are currently planning a large fMRI study in which we will be acquiring detailed information about the properties of the scanners we will be using during the study, by acquiring phantom data at regular intervals. We have been thinking about ways to exploit this information during the analysis of the resulting data, and have been discussing the Bayesian framework espoused within the FSL (and SPM) packages as a possible approach to incorporating this information.
From the phantom data, we are hoping to be able to estimate aspects of the noise properties of the scanner - for example, the variance of the observed signal, and perhaps also a characterization of its distribution. Whilst we could use this information at the 2nd level analysis, it would seem something of a waste – perhaps this information could be used in the formulation of a prior distribution, against which the explanatory power of the hypothesized model (reflected in the experiment design matrix) could be assessed at the first level of analysis.
Whilst Bayesian methods for fMRI data analysis have received substantial treatment within the literature, I have not found any reference to a scheme as outlined above. Perhaps this is because data about the noise properties of the scanner are not routinely taken, or because such an approach is excessively assumption-laden, or otherwise flawed or impractical. Any pointers, regarding previous literature, appropriate software, or known or supposed problems with the approach, would be most gratefully received.
Thank you in advance,
Henry Chase
|