Inheritance diagram for nipy.algorithms.group.parcel_analysis:
Parcel-based group analysis of multi-subject image data.
Routines implementing Bayesian inference on group-level effects assumed to be constant within given brain parcels. The model accounts for both estimation errors and localization uncertainty in reference space of first-level images.
See:
Keller, Merlin et al (2008). Dealing with Spatial Normalization Errors in fMRI Group Inference using Hierarchical Modeling. Statistica Sinica; 18(4).
Keller, Merlin et al (2009). Anatomically Informed Bayesian Model Selection for fMRI Group Data Analysis. In MICCAI‘09, Lecture Notes in Computer Science; 5762:450–457.
Roche, Alexis (2012). OHBM‘12 talk, slides at: https://sites.google.com/site/alexisroche/slides/Talk_Beijing12.pdf
Bases: object
Bayesian parcel-based analysis.
Given a sequence of independent images registered to a common space (for instance, a set of contrast images from a first-level fMRI analysis), perform a second-level analysis assuming constant effects throughout parcels defined from a given label image in reference space. Specifically, a model of the following form is assumed:
Y = X * beta + variability,
where Y denotes the input image sequence, X is a design matrix, and beta are parcel-wise parameter vectors. The algorithm computes the Bayesian posterior probability of beta in each parcel using an expectation propagation scheme.
Parameters: | con_imgs: sequence of nipy-like images :
parcel_img: nipy-like image :
parcel_info: sequence of arrays, optional :
msk_img: nipy-like image, optional :
vcon_imgs: sequece of nipy-like images, optional :
design_matrix: array, optional :
cvect: array, optional :
fwhm: float, optional :
smooth_method: str, optional :
res_path: str, optional :
write_smoothed_images: bool, optional :
|
---|
Save parcel analysis information in NPZ file.
Compute parcel-based posterior contrast means and positive contrast probabilities.
Parameters: | full_res: boolean :
|
---|---|
Returns: | pmap_mu_img: nipy image :
pmap_prob_img: nipy image :
|
Compute voxel-wise t-statistic map. This map is different from what you would get from an SPM-style mass univariate analysis because the method accounts for both spatial uncertainty in reference space and possibly errors on first-level inputs (if variance images are provided).
Returns: | tmap_img: nipy image :
|
---|
Helper function for Bayesian parcel-based analysis.
Given a sequence of independent images registered to a common space (for instance, a set of contrast images from a first-level fMRI analysis), perform a second-level analysis assuming constant effects throughout parcels defined from a given label image in reference space. Specifically, a model of the following form is assumed:
Y = X * beta + variability,
where Y denotes the input image sequence, X is a design matrix, and beta are parcel-wise parameter vectors. The algorithm computes the Bayesian posterior probability of cvect’*beta, where cvect is a given contrast vector, in each parcel using an expectation propagation scheme.
Parameters: | con_imgs: sequence of nipy-like images :
parcel_img: nipy-like image :
msk_img: nipy-like image, optional :
vcon_imgs: sequece of nipy-like images, optional :
design_matrix: array, optional :
cvect: array, optional :
fwhm: float, optional :
smooth_method: str, optional :
res_path: str, optional :
|
---|---|
Returns: | pmap_mu_img: nipy image :
pmap_prob_img: nipy image :
|