WAIR 2.0 Simple 2D Warp Movie

The goal of this example is to shed light on the problems researchers in the field of medical imaging, and in particular, brain mapping, face by trying to construct accurate anatomical and/or functional atlases of various human organs, and then using these atlases to study real data which may not a priori fit perfectly within the frame of the atlas. Suppose we want to register the image on the left to the atlas (middle), representing an average of a family of normal control subjects. Both images, data (left) and atlas (middle), represent transverse (axial) slices through the 3D volumetric data sets. Having the data (left) aligned into atlas space (middle) will allow us to use the comprehensive brain mapping associated with the atlas. Constructing such mapping is a very intensive process requiring many human expert hours in identifying structural regions, functional connectivity and structure-function associations.

The above movie illustrates the main three factors in any image registration technique: rotation, shift and local (non-linear) deformation. Warping, or image registration, is the process of aligning one curve, surface, volume, or higher dimensional manifold, to another object of the corresponding type. The goal of warping is to be able to construct a meaningful atlas representing the object of interest. The most important characteristic of image registration is the trade-off between goodness of fit and the amount of object-specific information retained in the residual resliced object following warping onto the atlas. On one hand we wish to align the data onto the atlas as well as we possibly can. A perfect registration, however, will obliterate the entire data information content of the object by making it virtually identical to the atlas. The registration vector field retains the collection of object-specific characteristics.

)