New to CliJ deconvolution

I want to try CliJ for deconvolving some images for a client, but the plugin asks for a convolution kernel. How do I develop a convolution kernel? Can I use a PSF image? The image was taken with a 10X objective with a 0.45 NA. Thank you.

1 Like

Hi @GopherConfocal,

yes, that parameter can be the PSF image. If it’s big, you might observe bad performance. We just discussed that in another thread and are working on a solution.

But thanks for trying. Would you mind letting us know how it goes?

Thanks!

Cheers,
Robert

1 Like

Hi Robert,

The deconvolution with CliJ went nowhere - I got a completely blank image. I have attached the image I want to deconvolve and a 2D PSF image I generated with the plugin PSF Generator with Born & Wolf Optical Model. I would appreciate whatever advice you have to offer. Thank you!

Mary743E5-S46-S4-GFP&NEST&HNA-10X-4CH-5Hoescht-BG.tif (2.8 MB) 743E5-S46-S4-GFP&NEST&HNA-10X-4CH_5PSF450nm2D.tif (257.8 KB)

1 Like

Hi @GopherConfocal,

sorry for the inconvenience. I tried with your images and can confirm the black result image. Obviously there is a bug in CLIJs deconvolution. However, if you convert your image to 32-bit before processing it, it should work. I will fix this bug with the next release. Thanks for reporting the bug btw! I :heart: feedback :slight_smile:

Furthermore, your PSF image is huge (256x256 pixels) and doesn’t contain much. You can crop out a center part, e.g. 33x33 pixels around the bright spot to speed up deconvolution.

By converting your input image to 32-bit and cropping the PSF to 33x33, I get this result after 10 iterations on my Intel GPU within just some seconds:

Could you please try again and let me know if it works?

Thanks!

Cheers,
Robert

2 Likes

Hi Robert,

Ah, I noticed that my PSF image was 32-bit while my Hoescht image was 16-bit. So having different bit depths doesn’t work for CliJ deconvolution? Converting my image to 32-bit did the trick. CliJ deconvolution worked pretty darn well! It pulled out edges very nicely in the very dense, central region of nuclei. Thanks a million!

Mary

743E5-S46-S4-GFP&NEST&HNA-10X-4CH-5Hoescht-BGCliJDeconRL.tif (5.5 MB)

1 Like

It’s supposed to work with different bit depths. And it will work with the next release. :smirk: Thanks again for trying and reporting the bug. You made CLIJ better today! :wink:

3 Likes