Added groupwise description with figure.

_pages/newmsm_guide.md

@@ -194,6 +194,10 @@ This will repeat Step 2 above, but this time each of the meshes will have a corr
 
 # Groupwise MSM
 
+We extended the existing MSM framework to perform groupwise registration by simultaneously co-registering clusters of surfaces, which share common modes of cortical variation. Let's say, there exist $n$ moving meshes for any given cluster; each input surface is therefore associated with its own low-resolution CP grid. For each iteration, the groupwise registration must choose the optimal discrete displacement for all CP grid by comparing the overlap of features, across all inputs, for the set of proposed moves. For each CP grid, this involves calculating the similarity between overlapping patches of features, defined around each of its CP vertices, and all overlapping patches defined for neighbouring vertices in all other CP grids. These are defined by propagating the proposed mapping to the resolution of the input feature maps, and resampling all data to a common reference frame ($F$) to calculate patch overlap. Here $F$ is defined from a high resolution icospheric tessellation (ico6 - 40,962 vertices), and features are resampled using adaptive barycentric interpolation. Regularisation is then implemented by calculating the hyper-elastic strain of every proposed move on all subjects' CP grid.
+
+<img src="../images/newmsm/gw_method_flow.png"/>
+
 # Configuration Files
 
 Configuration files modify all tunable parameters of the registration. For a full list of all registration parameters you can enter: