Dear Torsten,

Thank you so much for your suggestion. I agree with you that we cannot expect the abdominal scans to be aligned with rigid body registration very well, and as you said, what I was trying was just to use rigid registration to get a close start for ease of the following non-rigid registration. The idea of registration on extracted structures should be helpful, I have actually tried that when using other registration tools (flirt, niftyreg), which worked pretty well. But it didn't help for irtk if I stayed with the same configuration that I used for the raw images.My registered results were just not right or even close (worst case: off by almost the vertical distance of the whole field of view, where most part of the registered volume is empty), which made me believe I was most possibly not using the tool correctly. My purpose of trying irtk is actually to see how its non-rigid registration works comparing to niftyreg and fnirt, both of which did not quite work for me in my application. But it seems that I got a problem at the very first step.

Here is my update. I firstly resampled all images into isotropic resolutions, i.e., 1mm x 1mm x 1mm. Then instead of using rigid, I tried affine registration, which gave me much more reasonable result, where at least I can tell that the whole body is basically registered to around the right position and orientation, which is all I expect at this stage.

Actually, when I registered raw images without resampling, the intermediate results looked pretty reasonable to me if I turned on the debug option.

Now I am encouraging, but not fully confident whether isotropic resampling of resolution (maybe plus affine registration) is the solution to my problem. Do you have any experience that the resolution does matter when using irtk? 

Thank you!
Zhoubing