Epi-Fluorescence Microscopy and Deconvolution

Here you can post pictures and videos to show others.
Post Reply
Message
Author
User avatar
RobBerdan
Posts: 124
Joined: Tue Nov 20, 2018 3:26 pm
Location: Calgary
Contact:

Epi-Fluorescence Microscopy and Deconvolution

#1 Post by RobBerdan » Mon Apr 27, 2020 8:41 pm

Flourescence microscopy often introduces out of focus blur and haze and I have been experimenting with various techniques to reduce this haze. I stained Prototists with Acridine orange and compared processing with Photoshop, Image J (free software) and a trial version of Huygens Deconvolution.

My results show that Photoshop was the best. In this article I also provide references as to how to add Fluorescence to a microscope using LED flash lights and provide a basic introduction to Fluorescence microscopy that I hope will be of interest.

https://www.canadiannaturephotographer. ... scopy.html

Below are a few images from the article:
Attachments
Rotifer Brachionus after Photoshop Deconvolution
Rotifer Brachionus after Photoshop Deconvolution
DSC_0257_photoshopDeconvolution.jpg (65.96 KiB) Viewed 7709 times
Rotifer Brachionus RAW image showing Haze
Rotifer Brachionus RAW image showing Haze
DSC_0257_rawfile.jpg (32.88 KiB) Viewed 7709 times
Nuapilius larvae
Nuapilius larvae
S133_fs92_100.jpg (40.9 KiB) Viewed 7709 times
Amoeba
Amoeba
FS184_187.jpg (32.47 KiB) Viewed 7709 times
Bdelloid rotifer
Bdelloid rotifer
S131_131_133_Panorama1.jpg (38.38 KiB) Viewed 7709 times

Hobbyst46
Posts: 4277
Joined: Mon Aug 21, 2017 9:02 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#2 Post by Hobbyst46 » Mon Apr 27, 2020 9:52 pm

A very nice article (in the link), and a comprehensive and easy to read description of fluorescence microscopy. Thanks for posting.

May I place some questions and notes.

1. note: In epi-fluorescence, the objective serves a dual purpose - a condenser as well, hence the dependence of brightness on the NA is actually higher than in other illumination modes.

2. question: how was the Cymbella diatom stained with acridine orange ? was it alive before staining ? how does the result compare with an image of the autofluorescence of the diatom (including both chlorophyls and xanthines) ?

3. note:AFAIK, it is possible to avoid photobleaching by gating the excitation, without adding inhibitor chemicals, although that needs some electronic devices. Admittedly I have used gating for still objects, not video movies.

4. question: would deconvolution improve phase contrast images, by removing the halos ?

Thanks in advance for answers.

User avatar
RobBerdan
Posts: 124
Joined: Tue Nov 20, 2018 3:26 pm
Location: Calgary
Contact:

Re: Epi-Fluorescence Microscopy and Deconvolution

#3 Post by RobBerdan » Tue Apr 28, 2020 1:58 am

I will try to answer your questions:

1.The dependance of NA on brightness in some publications is to the 4th power, others to the square - the important thing is that higher NA results in much brighter images which is why I have been attempting to purchase objectives with the highest NA I can afford. I posted a question: How does a Flour objective compare to an Apochromat of the same NA in brightness? No one seems to have tested one against the other and replied. I have tested 20X objectives, but the Apochromat had a lower NA 0.65 and Flour 0.75 so I could not make a direct comparison. Most folks seem to think the Flour would be brighter and I tend to agree but don't know for sure.

2. The Diatom pictured is dead and was just in the debris that I stained with Acridine orange, no internal cellular components or chlorophyl. Same with the nauplius larvae photo - it was just the chitin exoskeleton.

3. Some microscopes do gate excitation or limit the exictation beam by moving it (confocal microscopes). I only used antifade agents on tissue sections using FITC-antibodies, otherwise I am looking at live specimens in water, though I am going to try 80% glycerol to see if the images might be clearer due to the increase in refractive index. See reference below about changing refractive index to remove halos. With diatoms we try to use high refractive index mountant e.g. Napthrax to increase the contrast. Good news is Acridine orange doesn't seem to fade much, however Fluorescein stains fade quickly in my experience.

4. Not sure if deconvolution would improve phase contrast I have not tried, but changing the refractive index can reduce halos - I have read they do this in some of the research papers I have read about yeast cells. I have also been searching the web, asking questions in research forums about software that mimic phase contrast and DIC which seems plausible, but none of the scientists that has published papers on this has responded to my questions strangely? It could be that they are being paid by microscope companies not to release their software, but it certainly seems like it would be possible to do.

The easiest remedy for removing or attenuating the intensity of halos is to modify the refractive index of the observation medium with higher refractive index components, such as glycerol, mannitol, dextran, or serum albumin - see https://www.microscopyu.com/tutorials/s ... -artifacts - for further information.

I hope this addresses your questions - Image J which does deconvolution is free to download, but takes some time to learn and it is widely used in research. I am just trying to get the best images I can out of my wide field fluorescent microscope as it is unlikely I will ever be able to afford a confocal microscope.

RB

Hobbyst46
Posts: 4277
Joined: Mon Aug 21, 2017 9:02 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#4 Post by Hobbyst46 » Tue Apr 28, 2020 6:53 am

Thanks RobBerdan. I will try to deconvolute images that I have collected with ImajJ.
For diatoms, the highest RI available is the choice anyway, for all illuminations.

User avatar
Wes
Posts: 1027
Joined: Sat Mar 09, 2019 12:58 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#5 Post by Wes » Tue Apr 28, 2020 7:15 am

Thanks a lot for the informative post. I will now record RAW images whenever I use fluorescence.
Zeiss Photomicroscope III BF/DF/Pol/Ph/DIC/FL/Jamin-Lebedeff
Youtube channel

Hobbyst46
Posts: 4277
Joined: Mon Aug 21, 2017 9:02 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#6 Post by Hobbyst46 » Wed Apr 29, 2020 8:58 pm

RobBerdan wrote:...
RobBerdan, which "blind deconvolution" plug-in or other tool do you suggest for the ImageJ ? I use ImageJ 1.52, and failed to find a deconvolution analysis tool, in my package, having never used this algorithm in the past.
Hobbyst46 wrote:
Tue Apr 28, 2020 6:53 am
Thanks RobBerdan. I will try to deconvolute images that I have collected with ImajJ.
For diatoms, the highest RI available is the choice anyway, for all illuminations.
Just found out that brightfield images are very difficult to deconvolve, in fact there is a fairly recent research about it. I conclude from it that phase contrast deconvolution is at least as difficult;

"Direct Imaging of Phase Objects Enables Conventional Deconvolution in Bright Field Light Microscopy"
Carmen Noemı´ Herna´ndez Candia1, Braulio Gutie´rrez-Medina2*, 2014.

On the other hand, reading that article made me think that perhaps darkfield images can be deconvolved, since they are somewhat similar to fluorescence. There are some articles about darkfield and deconvolution, from 2013.

User avatar
RobBerdan
Posts: 124
Joined: Tue Nov 20, 2018 3:26 pm
Location: Calgary
Contact:

Re: Epi-Fluorescence Microscopy and Deconvolution

#7 Post by RobBerdan » Thu Apr 30, 2020 3:11 am

Hi to do deconvolution in Image J you need to get a plug in and download - here is the link, read how to install the plugin -it is straight forward - https://imagej.net/Deconvolution

I have only seen deconvolution used for fluorescence and confocal microscopy. On photomacrography.net Rogelio Moreno used some deconvolution software (expensive software) and has very impressive results. https://www.photomacrography.net/forum/ ... hp?t=21618 - his deconvoluted images are some of the best I have seen.

I also found that Nikon NIS elements has the ability to do convolution- and even allows anyone to upload a file and they will produce the deconvoluted image for you that you can download on their web site - I just did a quick test yesterday and plant to test it further. https://deconv.laboratory-imaging.com/process - not as good as Photoshop but good.
I am looking into the NIkon NIS software to see how much it is and how good it is.

Cheers
RB

User avatar
Wes
Posts: 1027
Joined: Sat Mar 09, 2019 12:58 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#8 Post by Wes » Sun May 03, 2020 1:44 pm

I found this lecture useful to get an idea of the mathematics behind deconvolution and around 36:30 another approach (ER-decon) is mentioned that looks quite promising.

Zeiss Photomicroscope III BF/DF/Pol/Ph/DIC/FL/Jamin-Lebedeff
Youtube channel

User avatar
RobBerdan
Posts: 124
Joined: Tue Nov 20, 2018 3:26 pm
Location: Calgary
Contact:

Re: Epi-Fluorescence Microscopy and Deconvolution

#9 Post by RobBerdan » Mon May 04, 2020 8:15 pm

I have been testing different deconvolution programs, some of them very expensive and so far I have been able to do as good, sometimes better using Photoshop. I have added some comparisions to my article. I also found that Nikon NIS elements offers a web site where you can drag and drop a .TIF file, enter some data and will process the image for you, for free in about a minute and you can download the image for those that might like to try it.

The web site: https://deconv.laboratory-imaging.com/process

I am also testing AutoQuant X3 which costs more then $10,000 see results below
Auto Quant iterations 10-40 and Photoschop I interation
Auto Quant iterations 10-40 and Photoschop I interation
AQ_differentIterationsPS1IT.jpg (84.55 KiB) Viewed 7357 times
Autoquant Deconvolution vs Photoshop
Autoquant Deconvolution vs Photoshop
cells_deconvolution.jpg (70.72 KiB) Viewed 7357 times

User avatar
Wes
Posts: 1027
Joined: Sat Mar 09, 2019 12:58 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#10 Post by Wes » Mon May 04, 2020 9:28 pm

Can you provide a bit more info on the Photoshop deconvolution process? How does it calculate the PSF in order to carry out the deconvolution calculations?
Zeiss Photomicroscope III BF/DF/Pol/Ph/DIC/FL/Jamin-Lebedeff
Youtube channel

BramHuntingNematodes
Posts: 1538
Joined: Tue Jan 21, 2020 1:29 am
Location: Georgia, USA

Re: Epi-Fluorescence Microscopy and Deconvolution

#11 Post by BramHuntingNematodes » Mon May 04, 2020 11:23 pm

This is an interesting topic for sure. G'MIC (free) also has an implementation (very simple) of the Richardson-Lucy algorithm that uses either a Gaussian or other type of blur as the PSF. It produces very promising sharpening with conventional photo stacking according to preliminary trials. It looks like Huygens is not optimized for attractive imaging and is pretty dismissive of the kind of bootstrapped PSF generated by G'MIC. I don't know how the PS one works, but often in practical statistics, blunt methods work embarrassingly well.
1942 Bausch and Lomb Series T Dynoptic, Custom Illumination

User avatar
wporter
Posts: 353
Joined: Wed Oct 21, 2015 10:18 pm
Location: United States

Re: Epi-Fluorescence Microscopy and Deconvolution

#12 Post by wporter » Tue May 05, 2020 1:08 am

There is another deconvolution program called Focus Magic, which has a PS plugin, although the plugin may differ somewhat from the standalone version. There is a free trial. It also does de-motion-blurring. I have used it a bit in astrophotography.

[Edit:]
I forgot to say that it is a Windows program (64-bit), but runs fine via Wine under Ubuntu Linux.
Last edited by wporter on Tue May 05, 2020 4:58 am, edited 1 time in total.

User avatar
RobBerdan
Posts: 124
Joined: Tue Nov 20, 2018 3:26 pm
Location: Calgary
Contact:

Re: Epi-Fluorescence Microscopy and Deconvolution

#13 Post by RobBerdan » Tue May 05, 2020 2:16 am

Greg Benz describes a technique for Deconvolution sharpening in Camera RAW. What I do is open the image in Photoshop Camera Raw and adjust the Black, clarity, Haze an texture sliders to improve clarity. I reduce the luminance noise to reduce grain. I may adjust the vibrance, and saturation a little and sometimes the overall exposure. I bring the image into Photoshop tweak the levels, and remove dirt and debris that is on the background or sensor. I might even open it again in Camera Raw and repeat a second iteration.

For low magnification work you don't really need a PSF (point spread function) and I have been testing several high end programs and free software like Image J which are able to estimate or calculate a theoretical point spread from the data and image called Blind Deconvolution. AutoQuant X3 (about $13,000) is able to do a pretty good job using Blind Deconvolution using an estimated PSF. I show some pictures and the more software I test the more I am convinced that Photoshop is faster and can do a better job sometimes. Photoshop may not be scientifically accurate but it significantly improves detail and can reduce out focus haze

I am still learning but to determine a PSF you need subresolution fluorescent beads and a focus stack. I found that small vial of subresolution fluorescent beads cost about $500 - if anyone knows of some for about $50\ml please let me know. Also I still using processing many single slice images. Photoshop is not good at aligning hazy images. I have tested short focus stacks, but my Nikon Optiphot only offers 1 um stacks (fine focus) which I have to do by hand. Automated focus and stacking systems that I have seen are very expensive and are able to do focus in z direction 0.2 um and often require special antivibration table.

Some of the software companies that offer Deconvolution are looking at offering packages on a yearly rental price as they know that only researchers with grants can afford their software prices. I understand that their software has a small specialized audience so they may have to charge these high prices to stay afloat. Image J (free program) requires splitting the image into 3 channels, to do deconvolution and then putting the channels back together. I liked AutoQuant because the software instructions were clear and I was running Deconvolutions in about 30 minutes after trying their program and my initial results were good - but I can't afford what they are charging and the more I test different programs the better I feel about processing in Photoshop. There are always subjective decisions in Photoshop and also in the high end deconvolution software. Maybe Photoshop will add Deconvolution components in the future since they are also used to process astronomical images. My main goals are to reduce haze, out of focus blurr in fluorescence images, and make the images clear and crisp. I am prepared to do more work to achieve this and will continue to follow this process.

I hope this is helpful to anyone doing fluorescence microscopy - in my article I also have links to the various programs, web sites and research papers. The Nikon NIS Elements site will do quick Deconvolutions by uploading TIF files on their web site. I tested Phase contrast not much improvement, I found some improvement in Darkfield microscopy.

Glad to see this topic is of interest to some - not many amateurs I know use a fluorescence microscope, perhaps more will when LED for Flourescence microscopy become more popular and cheaper.
Cheers and thanks for your interest and questions.
RB

tocharian
Posts: 7
Joined: Thu Nov 29, 2018 12:59 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#14 Post by tocharian » Thu Apr 27, 2023 9:32 am

RobBerdan wrote:
Thu Apr 30, 2020 3:11 am

I also found that Nikon NIS elements has the ability to do convolution- and even allows anyone to upload a file and they will produce the deconvoluted image for you that you can download on their web site - I just did a quick test yesterday and plant to test it further. https://deconv.laboratory-imaging.com/process - not as good as Photoshop but good.
I am looking into the NIkon NIS software to see how much it is and how good it is.

Cheers
RB
I'm asked to input some parameters when using Nikon's online deconvolution service. Not sure what "Calibration" means.

Sure Squintsalot
Posts: 393
Joined: Mon May 16, 2022 3:44 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#15 Post by Sure Squintsalot » Fri Apr 28, 2023 4:24 am

Call me a numbskull, but I don't understand how you'd deconvolve an image without having a solid grip on how the image was convoluted in the first place. I guess that why you'd (have to) figure out the point spread function of your optical system and then feed that into a specifically designed de-convolution algorithm, of which there are maybe a half dozen(?)

Calling the results of Photoshop's de-blurring and contrast enhancement tools similar to a deconvolution algorithm's based on a known PSF is like calling a bicycle similar to an F-16 because they both result in net movement of a person; sure, both images may produce sharper images, but I'd bet dollars to donuts that what's sharpened is likely very different in each image. I suspect that for amateur purposes, photoshop post-processing is just fine, but I'm not sure we should be calling it "deconvolution".

BramHuntingNematodes
Posts: 1538
Joined: Tue Jan 21, 2020 1:29 am
Location: Georgia, USA

Re: Epi-Fluorescence Microscopy and Deconvolution

#16 Post by BramHuntingNematodes » Wed May 03, 2023 12:16 am

Sure Squintsalot wrote:
Fri Apr 28, 2023 4:24 am
Call me a numbskull, but I don't understand how you'd deconvolve an image without having a solid grip on how the image was convoluted in the first place. I guess that why you'd (have to) figure out the point spread function of your optical system and then feed that into a specifically designed de-convolution algorithm, of which there are maybe a half dozen(?)
I wouldn't go so far as to call you a numbskull but "blind deconvolution" has been an area of mathematical research for decades. It's less efficient than non-blind deconvolution, but whether or not the PSF is known a priori is not what distinguishes deconvolution from other methods of sharpening digital images.
1942 Bausch and Lomb Series T Dynoptic, Custom Illumination

DonSchaeffer
Posts: 3275
Joined: Sun Mar 22, 2020 10:06 am
Location: Winnipeg, Manitoba, Canada

Re: Epi-Fluorescence Microscopy and Deconvolution

#17 Post by DonSchaeffer » Wed May 03, 2023 2:41 pm

spectacular!

Sure Squintsalot
Posts: 393
Joined: Mon May 16, 2022 3:44 pm

Re: Epi-Fluorescence Microscopy and Deconvolution

#18 Post by Sure Squintsalot » Fri May 05, 2023 4:30 am

BramHuntingNematodes wrote:
Wed May 03, 2023 12:16 am
... whether or not the PSF is known a priori is not what distinguishes deconvolution from other methods of sharpening digital images.
I'd suggest that knowing the PSF a priori is precisely what distinguishes deconvolution from other methods of sharpening digital images. In fact, even in seismology, for which the technique was developed, deconvolution absolutely requires a priori knowledge of signal behavior through materials. As for how PhotoShop sharpens images, at least my version uses the "nearest neighbor" method:
Screenshot 2023-05-04 214052.jpg
Screenshot 2023-05-04 214052.jpg (92.31 KiB) Viewed 2330 times
For sure, the nearest neighbor algorithm is computationally intensive and produces pleasing results, but no photonics engineer will call it a deconvolution algorithm. In fact, an exhaustive 4.863 minute google search finds no legitimate association of the terms "Photoshop", "deblurring", and "deconvolution".

There's probably an excellent reason for why biologists do NOT sharpen their fluorescence images with Photoshop. Sharpening a landscape photo, portrait photo, picnic photo, sports photo, or wedding photo to pleasing effect is one thing, but quantitative imaging of fluorescence emitting subjects, laser illuminated objects, or an astronomical bodies is an entirely different beast.

BramHuntingNematodes
Posts: 1538
Joined: Tue Jan 21, 2020 1:29 am
Location: Georgia, USA

Re: Epi-Fluorescence Microscopy and Deconvolution

#19 Post by BramHuntingNematodes » Fri May 05, 2023 12:43 pm

I am not immediately familiar with Photoshop, but I assume when Berdan said he was using Photoshop deconvolution he was using some implementation of a deconvolution algorithm in Photoshop, not one of photoshops non-deconvolution algorithms.

I would argue that a method that estimates a psf just from the image data and deconvolutes assuming that psf is still deconvolution.
1942 Bausch and Lomb Series T Dynoptic, Custom Illumination

BramHuntingNematodes
Posts: 1538
Joined: Tue Jan 21, 2020 1:29 am
Location: Georgia, USA

Re: Epi-Fluorescence Microscopy and Deconvolution

#20 Post by BramHuntingNematodes » Fri May 05, 2023 12:48 pm

I guess I am confused, does Photoshop have a deconvolution algorithm? I assumed that was what was being discussed as gnu image manipulation program has a genuine one and it is merely a free clone of whatever version Photoshop was at a few years ago.
1942 Bausch and Lomb Series T Dynoptic, Custom Illumination

Post Reply