Probabilistic Functional Maps


Probabilistic Functional Maps

When performing fMRI group studies, it is very interesting to investigate the spatial consistency of activity patterns across subjects. One useful approach to quantify this consists in the calculation of probabilistic functional maps. At each spatial location, such maps represent the relative number of subjects leading to significant task activity. In a study of 20 subjects, for example, a value of 60% would mean that 12 subjects activated the respective brain region. It is evident that the calculation of probabilistic functional maps depends to some extent on the chosen brain normalization method since probabilities are determined by counting how many subjects do activate at the “same” spatial location. For volumetric normalization schemes, spatial coordinates (e.g. in Talairach or MNI template space) address homologue regions in the brains of different subjects. For cortex-based normalization schemes, aligned surface points (vertices) are used to address homologue brain regions. Relative to a good macro-anatomical alignment, probabilistic functional maps may reveal the spatial variability of functional brain areas.

Probabilistic Functional Maps - 2
The calculation of probabilistic functional maps is one of the new features of release 1.9,12 of BrainVoyager QX. Probabilistic maps can be calculated based on regions-of-interests specified for each subject in volume (VOIs) or surface (POIs) space as well as from subject’s activation maps in volume (VMPs) and surface (SMPs) space. Using ROIs has the advantage that homologue functional regions have been already identified for each subject. Probabilistic maps on the basis of subject-specific contrast maps may lead to less focal results than the ROI approach, but it may help to reveal unpredicted consistent regions or extended networks across subjects.
Prior to the new release, the variability of ROIs could be visualized by showing VOIs or POIs from different subjects in different colors. The snapshot above shows this approach for the “extrastriate body area” (EBA) as identified in 12 subjects in Talairach space. Such a display does, however, not reveal how many subjects overlap at each voxel or vertex. In order to get this information, the new probabilistic map ROI tools can be used.

Probabilistic Functional Maps - 3
The sanpshot above shows the new “VOI Probabilistic Map Creation” dialog, which can be called from the “VOI Functions” tab of the “VOI Options” dialog. The program needs as input VOIs with names identifying the subject ID part and the ROI part. As usual, the subject ID can be defined either as the first (as in this example) or last part (see “Naming convention” field).

Probabilistic Functional Maps - 4
In the snapshot above, the volume map was created with the same resolution as the anatomical VMR data set (“Map resolution” parameter “1x1x1”). It is also possible to create the probabilistic maps in coarser resolutions with respect to the VMR data set by using the “2x2x2” or “3x3x3” option. Using the same VOI data, the probabilistic map with “3x3x3” resolution is shown in the snapshot below.

Probabilistic Functional Maps - 5
The sanpshot below shows the corresponding dialog for the creation of POI-based probabilistic maps. In this case, POIs are defined on individual meshes used for cortex-based alignment (“SPH” versions of cortex meshes).To exploit the macro-anatomical alignment of gyri and sulci, the POIs from a single subject has to be mapped to the common space using the “SSM” files produced during cortex-based alignment of the subject population. This information can be specified in the “Subjects with cortex-aligned info” table, which maps the subject ID strings to the respective SSM file. If this table has been created once for a group of aligned brains (usually one set per hemisphere), the table can be saved to disk for later usage in the new “S2S” (Subject-ID to SSM) file.

Probabilistic Functional Maps - 6
An example of a resulting probabilistic map for three POIs (FFA, PPA and LOC) is shown on the top of this blog.

Probabilistic Functional Maps - 7
As stated above, probabilistic maps can also be calculated from subject-specific functional maps. As input, the same contrast map must be available for each subject in the “Volume Maps” or “Surface Maps” dialog. The set of subject-specific maps can be easily created with the “” feature of the “Overlay GLM Options” dialog. Note that for the creation of probabilistic maps, only positive t values are used; to get, for example, a probability map for face selective areas, the contrast “Faces > Houses” might be used; to get a probabilistic map for “house-seelctive” areas, the opposite contrast (“Houses > Faces” ) might be used. Besides selecting an appropriate contrast depending on a certain experimental study, the outcome of the probability map will depend also on the thresholds used for each map: A subject will be counted as “activating” at a spatial location, if the map value at this location passes the maps threshold. While thresholds should be the same across subjects, it might sometimes be justified to use different threhsolds per subject.

Probabilistic Functional Maps - 8
For volume space, probabilistic maps can be calculated by using the “Create” button in the “Probabilistic maps” field of the “Volume Maps Options” dialog. An example of a volume-based probabilistic map is shown in the snapshot above (Five subjects, contrast “Faces > Houses”).
The result of a surface-based probabilistic map is shown below (same data, contrast “Faces > Houses” (green) and contrast “Houses > Faces” (blue)).

Probabilistic Functional Maps - 9

More details about probabilistic map creation can be found in the updated User’s Guide of the 1.9.12 release, which should be available from our web site in a few days. The User’s Guide also describes additional enhancements of the program update including new options for high-pass filtering and design matrix creation.