Elastix on fluorescence microscopy data

[DONATO SARDELLA]

Hi everyone,

I am trying to register 3D stacks acquired with fluorescence microscopy over time. Since the samples are observed at multiple time points, misalignment errors and structural changes in the sample can be significant (e.g. shifts and rotations, some structures moving within the general stack).

Data has a submicron voxel size with a lot of fine detail that changes over time and so far I have been able to sporadically get somewhat OK registrations by downsampling my data significantly before starting the registration (xy downsampling factor 10X for xy and 3X for z) using 10k samples (10% of total) and more than 10-20k iterations. At this level of downscaling my volumes have a pixel size of 150x150x50px instead of the original 1500x1500x150.

I am using the elastix wrapper in FIJI so my parameter file includes smoothingpyramids, Mattes mutual information and adaptive stochastic gradient. It feels like the optimizer is converging too slowly since a smaller number of iterations fails to sufficiently translate/rotate the image (Euler transform).
I have also played around with starting the program from the command line and using different metrics/optimizers without much success.

I'm open to any suggestion you may have. Thanks!

PS I have used initial transformations obtained in FIJI's Big Warp by placing manual landmarks and it is also not working well.

[Shane Kirkbride]

Is this data rigid or non-rigid data?

-Shane

--
You received this message because you are subscribed to the Google Groups "elastix-imageregistration" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elastix-imageregis...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/elastix-imageregistration/b9ea615a-d113-4634-b1ca-adbd31d59dc5n%40googlegroups.com.

[DONATO SARDELLA]

Apologies for the delay, I must have failed to hit the send button and then closed the tab with my reply.
The data is non rigid in the sense that I'm scanning soft kidney tissue without any rigid structure(e.g. bones), but each volume at each time point can be thought of as a rigid block without need to register the single slices (I take care of that with another algorithm if necessary). I am able to perform manual registration with multiple manual landmarks using the relative positions of soft structures (renal tubules) that remain relatively stable over time.