The authors present a framework embedded in ITK for optimizing an unstationary velocity field for doing pairwise image registration. The code uses several classes that are part of the ITKv4 registration framework. Specifically, they consider the case of a velocity field represented by a BSpline field. 

The code presented here is generic as it support any image to image metric in ITK.

Authors give qualitative examples of image warping and program usage in their paper. 

The code is fully open source and part of ITK.

The experiments would be easier to reproduce if the input images could be added directly downloaded from the insight journal page. 

ITK is the only open source package needed here. 

The submission comes with examples about how to use the command line programs of this submission. 

The code is well commented and up to the highest ITK standards. 

I could not find the data. 

I would encourage the authors to think about the extension of such framework to handle sequences of images rather than pair of images.

I am convinced that a time varying representation of the transformation has a special interest when solving for time varying transformations e.g. evolution of a tumor over time, hear deformation over one cycle etc ... 

The current implementation requires to fully resample the fixed and moving image volumes at each iteration. This can be quite time consuming. It would be interesting to disccuss alternatives for avoiding this resampling.

In case a paper describing the mathematical framework behind this implementation exists, it would be great to add it to the references.