The MDSS plugin works on both volume and surface maps. Depending on the currently active document (a VMR or an SRF), the plugin will either consider a series of VMP loaded (or overlaid) to the current VMR, or a series of SMP loaded (or overlaid) to the current SRF. If no maps are overlaid, the MDSS plugin GUI dialog will start from the first tab. The voxel or vertex selection determining the actual number of dimensions of the MDS analysis can be performed using VTC mask or VOI files in the volume space or POI files in the surface space.
Note: When using a mask-based selection in the volume space, it is important to select a mask file which is consistent with the VMP file, i. e. the VTC mask should correspond to the VMP in terms of the internal bounding box and voxel resolution. If this turns out to be not the case the calculations will not return any result. In contrast, no special attention should be paid by the user when using VOI files. In fact, these are intrisically defined on the anatomical space and the plugin will internally generate the proper mask for the VMPs. In the surface space, the POI files must be consistent with both the mesh (SRF) and the map (SMP) space, in the sense that the number and order of vertices should correspond. On the other hand, as far as both the surface maps are generated, and the POI files defined, on the current mesh, there is no risk of inconsistent dimension selection.
When calling the MDSS plugin, when the series of VMP (SMP) is already overlaid to the currently active VMR (SRF), the GUI dialog will automatically switch to the second tab called "Class selection". If not, the maps are to be loaded to the first tab. To be considered as valid input for the MDSS plugin, the series of map should comply with one of the following naming conventions for the single maps:
Convention 1 (multi-study GLM convention, K studies, N contrasts):
<Study 1: Ctr 1>Convention 2 (MVPA convention, N classes, K trials):
<Class-1_Trial-1>Convention 3 (multi-subject GLM convention, K subjects, N contrast):
<Subject XX: Ctr 1>Convention 4 (sog-ICA convention, K clusters, N studies/subjects):
<Study/Subject XX: Cluster 1>Note: this naming convention is automatically followed by the program tools and plugins (e. g. MVPA, GLM, ICA) that are able to generate as series of maps with this labeling of the maps. For instance, from the Overlay GLM, with a multi-study GLM file loaded, it is possible to first create a number of contrasts and then generate the corresponding maps (beta maps or t-maps) with the appropriate naming from the options of this dialog, as shown in the screenshots below.
Once a VMP or SMP is loaded or overlaid, the MDSS plugin will strictly follow the naming convention to label the maps and allow the grouping of the maps based on the class or group membership. The fundamental distinction between classes and groups is that classes usually pertain to a within-subject factor (e. g. a stimulus) and the groups pertain to a between-subject factor (e. g. a specific population). As a further alternative, it is possible to use classes and groups to respectively label trials and stimuli in the context of MVPA. The second and third tabs allow to (un)check the class labelling and assign/activate the group labelling which is originally derived from the map naming, as illustrated in the next screenshots:
Even if no class or group is selected/assigned, MDS is anyway applied to all maps and all points are plotted "unclassified", i. e. without labeling or grouping. Thereby, the user can visualize the point distribution even without membership information. When classes or groups are assigned (and selected) one or two additional plots are generated containing the labeling information respectively coded as color or category of the projected points, as illustraed in the next screenshots for a single-subject multi-study GLM analysis with 5 studies and 2 conditions included in the contrasts as classes (naming convention: 1). Here the MDS is applied to all voxels from a selected VOI:
The MDSS plugin can be applied to surface maps when the current document is an SRF mesh loaded to the surface module. In the next screenshots, an example is given where the MDSS is applied to a series of maps from a multi-subject GLM (naming convention: 3), with 10 subjects and four conditions. For a given POI, the plots are generated respectively for 2 classes (corresponding to conditions "Faces" and "Houses") and 2 groups (corresponding to the first and last five subjects of the groups):
As anticipated, the MDSS plugin can also be applied to MVPA and SOGICA generated series of maps (naming convention 2 and 4). The next two screenshots illustrate these two cases obtained after an MVPA (naming convention: 1) and SOGICA (naming convention: 4) analysis:
Finally, the last tab of the MDSS plugin gathers a number of advanced options to change: (i) the way the original distance matrix is obtained from the similarities (spatial correlations) and (ii) the projection type from the (default) linear MMDS to a non-linear type (Sammon or CCA). Using the square root (and eventually increasing the value of p) in the similarity-to-dissimilarity transformation has the effect to make smaller distances correspond to more dense projections and higher distances to less dense projections. As a result, the distribution of the distances spreads in such a way that possible compact clusters in this distribution can be seen better, as shown in the next screenshots depicting two scenarios with p=1 (default) and p=5.