Dear Glad,
The spatial priors in the Bayesian routines in SPM are not priors over the location of activations. They are rather priors that constrain the estimates of regressions coefficients at a voxel to be similar to those at nearby voxels.
However, what you describe is a "functional localiser" and can be implemented in a number of ways in SPM. For classical inference one can write out localiser images (e.g. t/F maps at a liberal threshold) and then constrain the results of a subsequent analysis using either e.g. an explicit (inclusive) mask, or Small Volume Correction (SVC). Or for Bayesian inference one can also use an explicit mask (specifying this at the estimation stage will greatly speed things up).
All the best,
Will.
PS. If you or other readers are familiar with M/EEG source reconstruction, an analogy here is one of Distributed Source reconstruction (local constraints) rather than one of Equivalent Current Dipioles (constraints over locations). Tom Nichols at Warwick, and others in the field, have been developing these latter ECD type models for fMRI but i'm not sure there's any software available.
________________________________________
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of Paul Glad Mihai <[log in to unmask]>
Sent: 17 December 2015 12:55
To: [log in to unmask]
Subject: [SPM] Bayesian inference and spatial priors
Dear SPM List,
I was wondering wondering about Bayesian estimation and spatial priors,
particularly can one use a functional activation map form another
experiment as a spatial prior during Bayesian estimation?
For example, in one experiment one would present a checkerboard pattern
to localize V1 then use this map as a hierarchical prior in the
estimation in a specific visual task experiment? Does this make sense?
Regards,
Glad
--
Paul Glad Mihai, PhD
Independent Research Group "Neural Mechanisms of Human Communication"
Max Planck Institute for Human Cognitive and Brain Sciences
Stephanstraße 1A, 04103 Leipzig, Germany
Phone: +49 (0) 341-9940-2478
E-mail: [log in to unmask]
|