[MINC-users] optizing minctracc parameters for heart data

[Jason Lerch]

>> One of the two data sets spans a good part of the chest, but the registration can probably be determined by the heart and aorta alone, and perhaps incrementally improved with masks based on the coronary arteries and aorta.
>>     
>
> Sounds suspiciously like what happens with registration of mouse brain
> data. Here I was hoping they would comment on this but no luck so far!
>   

Allright, I'll bite. We never scale the organ, we change the parameters 
instead - and -w_translations is the one that you want. We've found it 
to be fairly insensitive to the exact values, i.e. -w_translations of 
0.2 0.2 0.2 works as well as 0.06 for a 0.06 voxel size. Occasional very 
low values cause segfaults - which we've never gotten around to 
debugging as they don't happen very often, but be aware. We don't 
usually change -w_rotations. As for step size, we always use the a 
series of decreasing step sizes, ending up at either voxel size or, more 
commonly, twice the voxel size. We use a -sub_lattice of 6 and 
-lattice_diameter of 3 times the step size in each minctracc call.

As for masking - that really depends on how variable the data outside 
the region of interest for you is. Low variability between source and 
target means that the mask won't have that great of an impact, high 
variability means that it could be essential. Similar answer with 
whether to use the gradients of the blur - for brains we always use them 
as that gives us better convergence (and mouse brains are very similar 
to each other), for something like mouse embryos the gradients makes the 
registration worse so we don't use it.

Hope this helps,

Jason

> :)
>
>
>
>