Fwd: Parallel Iterative Deconvolution

[Daniel James White]

Hi Esteban and all,

I have been testing Piotr W's deconv plugins this week
with some interesting results.

I used the versions (1.9) from the link on the Fiji wiki page for
parallel and spectral iterative deconvolution.
I just put the .jars in the Fiji plugins dir manually.

I made PSFs with the Diffraction PSF 3D  plugin from  Bob D
and convolved ground truth images with the Convolve 3D plugin also  
from Bob D.
I also tested Bobs Iterative Deconvlove 3D plugin (runs on a single  
thread)

With the bat cochlea volume sample data set,
convolved with a theoretical PSF, added gaussian noise, then  
deconvolved with the same PSF,
i got the most sensible results using the
Parallel Iterative Deconvolution
WPL method (100 max iterations, single precision, reflexive  
boundary .... default settings)
This looked very similar to the result from Bob's Iterative Deconvolve  
3D,
as one might expect as its pretty much the same method, just  
parallelised.

On a 8 core machine it's pretty fast.....

I also got a reasonable result with the
Parallel Spectral Deconvolution
Generalized Tikhonov (reflexive), Laplacian,  threshold 0, auto  
regularization.
This gave a result with possibly more artifacts (only low intensities,  
nothing really really bad)
but is super super fast on a machine with 8 cores.
More time is spent in the setup and pre calculation steps that the  
actual doconv it seems.
Its a bit quick and dirty maybe, but sometimes thats the best you can  
do anyway (noisy data?)
and it might be good enough for some tasks...(not intensity analysis  
perhaps)

On real fluorescence microscopy data I also got reasonable results  
using these methods.
Note the default MRNSD method didnt seem to do as good a job
on my ground truth data as the WPL method.
Maybe there is more to be investigated there.

Note that Bobs diffraction psf 3d plugin generates widefield PSFs only.
For a confocal PSF... well you can approximate it by making the square  
of a widefield one... but i'm not sure how good that is.
There is a thread about that currently running on the ImageJ list...  
subject =  Confocal PSF (was Puncta quantitation)

Generally,  measured PSF is nearly always the best (but sometimes  
overkill for very noisy data?)
Lenses and other optics are often far from perfect and a theoretical  
PSF may be way off the real one of the system.

cheers

Dan

> "G. Esteban Fernandez" <[hidden email]> Apr 05  
> 11:10AM -0700 ^
>
> Hello,
>
> I just installed Fiji primarily to use the Parallel Iterative
> Deconvolution plugin that it implements, but I don’t see it anywhere.
> Can someone help me out please?
>
> Thanks,
> Esteban
>
> --
> G. Esteban Fernandez, Ph.D.
> Imaging Scientist
>
> Cellular Imaging Core
> Childrens Hospital Los Angeles

Dr. Daniel James White BSc. (Hons.) PhD
Senior Microscopist / Image Visualisation, Processing and Analysis
Light Microscopy and Image Processing Facilities
Max Planck Institute of Molecular Cell Biology and Genetics
Pfotenhauerstrasse 108
01307 DRESDEN
Germany

[phone] (German Mobile)
[phone] (Work phone at MPI-CBG)
[phone] (Fax MPI-CBG LMF)

http://www.bioimagexd.net  BioImageXD
http://pacific.mpi-cbg.de                Fiji -  is just ImageJ (Batteries Included)
http://www.chalkie.org.uk                Dan's Homepages
https://ifn.mpi-cbg.de  Dresden Imaging Facility Network
dan (at) chalkie.org.uk
( white (at) mpi-cbg.de )