Fibroblast cells with and without deconvolution
Moderators: rjlittlefield, ChrisR, Chris S., Pau
-
- Posts: 1511
- Joined: Mon Jan 15, 2018 9:23 pm
- Contact:
Fibroblast cells with and without deconvolution
Improved the acquisition workflow by a lot, fixed that annoying green hue which was due to me being silly.
Nucleus: shortpass DAPI
Mitochondria and actin: dual band FITC and TRITC, separated with modulator
Three total stacks.
Widefield fluorescence:
Deconvolved:
Nucleus: shortpass DAPI
Mitochondria and actin: dual band FITC and TRITC, separated with modulator
Three total stacks.
Widefield fluorescence:
Deconvolved:
Re: Fibroblast cells with and without deconvolution
Wow! That's really nice!
-
- Posts: 1617
- Joined: Sat Apr 14, 2018 10:26 am
Re: Fibroblast cells with and without deconvolution
These are some beautiful results! The deconvolution is particularly interesting--have you written about what's involved with that before?
-
- Posts: 1511
- Joined: Mon Jan 15, 2018 9:23 pm
- Contact:
Re: Fibroblast cells with and without deconvolution
Certainly worth an article on that! I come from mathematics and electrical engineering background, so I know the theories behind it well enough to put out something for the layman.Scarodactyl wrote: ↑Fri Aug 05, 2022 8:08 amThese are some beautiful results! The deconvolution is particularly interesting--have you written about what's involved with that before?
However, it is not worth the effort in my opinion, considering how expensive software is and how much time it takes.
-
- Posts: 1617
- Joined: Sat Apr 14, 2018 10:26 am
Re: Fibroblast cells with and without deconvolution
How similar are the results vs unsophisticated sharpening?
-
- Posts: 1511
- Joined: Mon Jan 15, 2018 9:23 pm
- Contact:
Re: Fibroblast cells with and without deconvolution
Those are incomparable, I may be doing a widefield + deconvolution versus confocal comparison soon.Scarodactyl wrote: ↑Mon Aug 08, 2022 9:19 amHow similar are the results vs unsophisticated sharpening?
The widefield image shown has underwent sharpening.
Re: Fibroblast cells with and without deconvolution
I don't know if this true but I read somewhere that the "Detail" slider on the sharpening panel of the previous version of Adobe Camera Raw toggles between an on-the-fly deconvolution (low values of slider) and unsharp mask (high values of slider). It is true that the halo artifacts characteristic of the unsharp mask algorithm disappear when the slider is set to its lowest value.
Re: Fibroblast cells with and without deconvolution
If someone wants to try deconvolution (and has matlab) at home there is a quite recent publication by Guo et al. "Rapid image deconvolution and multiview fusion for optical microscopy" https://www.nature.com/articles/s41587-020-0560-x which also offers ready to use code. Compared to the previously available free agorithms in imagej its super fast and produces good results even with a computed psf.
Its not perfectly free but matlab licenses are more widespread than huygens so maybe this is of some help.
Its not perfectly free but matlab licenses are more widespread than huygens so maybe this is of some help.
-
- Posts: 1511
- Joined: Mon Jan 15, 2018 9:23 pm
- Contact:
Re: Fibroblast cells with and without deconvolution
Yes, that is true.Lou Jost wrote: ↑Tue Aug 09, 2022 8:09 amI don't know if this true but I read somewhere that the "Detail" slider on the sharpening panel of the previous version of Adobe Camera Raw toggles between an on-the-fly deconvolution (low values of slider) and unsharp mask (high values of slider). It is true that the halo artifacts characteristic of the unsharp mask algorithm disappear when the slider is set to its lowest value.
With Photoshop and Capture One, if you set the detail slider to minimum (0.1 in Photoshop's smart sharpening console and 0.2 in Capture One), then drag the sharpen slider to something outrageous, it kind of works like deconvolution.
I do not how it is technically done, but how would Photoshop know the NA of my objective? The software I use requires the input of NA and other parameters.
- rjlittlefield
- Site Admin
- Posts: 23561
- Joined: Tue Aug 01, 2006 8:34 am
- Location: Richland, Washington State, USA
- Contact:
Re: Fibroblast cells with and without deconvolution
I assume that your software is using NA and the other parameters to calculate an estimated PSF.Macro_Cosmos wrote: ↑Sun Aug 14, 2022 2:37 amhow would Photoshop know the NA of my objective? The software I use requires the input of NA and other parameters.
There are other schemes that require imaging known targets such as tiny fluorescent beads, and then the PSF can be calculated from what the images look like.
Photoshop does not know the NA etc, but you tell it directly with the Radius slider to assume a certain width of the PSF . The overall process is essentially blind deconvolution, with you the human controlling the search for best PSF based on the image that results from filtering. I generally describe this as "sharpening to taste", since there is never an explicit criterion and different people will disagree about what is best.
--Rik
Re: Fibroblast cells with and without deconvolution
The interesting thing about this Photoshop method is that it does not give haloes, very unlike the behavior of Unsharp Mask. Another interesting thing about it is that Adobe has eliminated it in their latest ACR versions, as they continue their slow dumbing down of Photoshop.
-
- Posts: 1511
- Joined: Mon Jan 15, 2018 9:23 pm
- Contact:
Re: Fibroblast cells with and without deconvolution
The software can use both. I use simulated PSFs for deconvolution. The beads I have are too large to acquire an experimental PSF -- it will be bloated and thus inappropriate. According to many articles I have read, the bead size depends on the microscopy method. Every researcher seems to have a favourite way to calculate the sizes and stand by it. Generally, 100-200nm beads are suitable for widefield fluorescence. Mine are either too large or too small. Moreover, it is recommended to mount the beads with accordance to the slides themselves. For example, my fibroblast cells were mounted in ProLong Diamond, so it is best practice to have my fluorescence beads mounted in the same manner with the same procedure and coverslip. It gets really esoteric and pedantic, but I can see why people are like that when we are discussing PSF engineering, haha.rjlittlefield wrote: ↑Sun Aug 14, 2022 10:48 amI assume that your software is using NA and the other parameters to calculate an estimated PSF.Macro_Cosmos wrote: ↑Sun Aug 14, 2022 2:37 amhow would Photoshop know the NA of my objective? The software I use requires the input of NA and other parameters.
There are other schemes that require imaging known targets such as tiny fluorescent beads, and then the PSF can be calculated from what the images look like.
Photoshop does not know the NA etc, but you tell it directly with the Radius slider to assume a certain width of the PSF . The overall process is essentially blind deconvolution, with you the human controlling the search for best PSF based on the image that results from filtering. I generally describe this as "sharpening to taste", since there is never an explicit criterion and different people will disagree about what is best.
--Rik
I have been wondering if it is possible to use ZS to obtain an experimental PSF. People generally use ImageJ.
Thanks for the information about Photoshop's sharpening, I think I now know why I get barely any improvements when attempting to "sharpen" already deconvolved images.
If I were to guess, perhaps using Unsharp Mask, a "larger radius" which Rik pointed to was used, and therefore the bloated "assumed PSF" creates silly false detail displaying as halo artefacts. One of the many downsides of deconvolution is creating false details. I have encountered that already. My Nikon Z6's fixed pattern readout noise was also deconvoled as well (usually eliminated by Nikon using processing wizardry, of course, not just Nikon has it, all CMOS sensors do), showcasing ugly fringe lines, that is another day's rant.Lou Jost wrote: ↑Sun Aug 14, 2022 11:36 amThe interesting thing about this Photoshop method is that it does not give haloes, very unlike the behavior of Unsharp Mask. Another interesting thing about it is that Adobe has eliminated it in their latest ACR versions, as they continue their slow dumbing down of Photoshop.
Re: Fibroblast cells with and without deconvolution
"If I were to guess, perhaps using Unsharp Mask, a "larger radius" which Rik pointed to was used, and therefore the bloated "assumed PSF" creates silly false detail displaying as halo artefacts."
No, Unsharp Mask is a totally different algorithm that doesn't use a psf. I think the Detail slider simply combines these two methods (convolution and USM) in proportion to the slider position; farther to the right= more USM, farther to the left = more convolution.
No, Unsharp Mask is a totally different algorithm that doesn't use a psf. I think the Detail slider simply combines these two methods (convolution and USM) in proportion to the slider position; farther to the right= more USM, farther to the left = more convolution.
- rjlittlefield
- Site Admin
- Posts: 23561
- Joined: Tue Aug 01, 2006 8:34 am
- Location: Richland, Washington State, USA
- Contact:
Re: Fibroblast cells with and without deconvolution
The terminology is different, but Unsharp Mask (and most other simple filtering methods) will exactly un-do some blurring operation. Whatever blurring operation that is, plays the same role as the PSF in deconvolution. See my explanation at https://www.photomacrography.net/forum/ ... 198#203198 .Lou Jost wrote: ↑Sun Aug 14, 2022 7:07 pm"If I were to guess, perhaps using Unsharp Mask, a "larger radius" which Rik pointed to was used, and therefore the bloated "assumed PSF" creates silly false detail displaying as halo artefacts."
No, Unsharp Mask is a totally different algorithm that doesn't use a psf. I think the Detail slider simply combines these two methods (convolution and USM) in proportion to the slider position; farther to the right= more USM, farther to the left = more convolution.
--Rik