Deconvolution for 2D brightfield microscopy

Theory
1

I am interested in playing with deconvolution/deblurring on diagnostic 2D brightfield microscopy images, with H&E staining. Frankly, it’s more aesthetics and curiosity than real investigation.

I have a little experience with doing this with confocal images. In confocal microscopy, you can buy flourescent beads to provide point sources and estimate the PSF from that. I tried dumping some beads into mounting medium and taking a peek without flourescence, but of course they were not point sources. Instead, each bead was a little spherical lens, and was unsuitable for determining the PSF.

I haven’t seen any published stuff on deconvolution of standard 2D lightfield microscopy images. That leads me to believe that it’s either so easy I’m missing something basic, so hard that it’s too difficult to waste time on, or uninteresting to most people. Is this something anybody does? Are there any sources about it? Any places that sell conventional slides with tiny black dots?

Thanks!

billo

2

Hello @billo and welcome to the forum!
The more you look the more you will find that there is actually quite an extensive published literature out there for brightfield decon. Not as much in volume as the fluorescence decon literature - that is partly because most applications in routine practice are for fluorescence decon, esp. 3D from confocal, etc. but it is also partly because it is actually a more complex problem to do decon in brightfield (bascially, fluo decon is easier / simpler to model in maths and hence to perform in practice than brightfield).

I have a special interest in decon and brightfield decon in particular. However, before you go any further, you need to define what you actually want to ‘fix’ with your deconvolution because that makes a lot of difference to how you go about it. For example, are you trying to ‘refocus’ an image? Or are you trying to remove some optical aberration like spherical or distortion in the 2D plane? Or are you trying to separate the various colours of the staining (digital de-staining)? Or are you trying to recover resolution beyond the Rayleigh ‘limit’? Or are you trying to remove focus blur from a 3D stack (you already stated that 3D was not your aim so I know that)?

In general using an emissive fluo bead will not give you the correct PSF for brightfield because there are other effects going on in bright field that complicate the model - particularly phase / scattering contrast in addition to absorption contrast (the latter is a kind of negative version of emissive contrast used by the fluo people).

I have developed special techniques for brightfield decon but my research in those fields were for 3D optical sectioning, stain separation and resolution recovery so they will not all be directly relevant to you but here are some of my papers in case you want to have a look:

This one explains a method of how to measure the brightfield PSF and gives a simple 3D optical sectioning example (not corrected for spherical aberration):
https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1365-2818.2009.03323.x

This one shows a practical application with some nice linked 3D fly-into videos of the interphase nucleus:

This one is about digital stain separation (not deconvolution exactly but related):
https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1365-2818.2010.03390.x

If you do a Google Scholar search for “Brightfield Deconvolution” you will come up with a lot more.
I am planning to do a tutorial series on deconvolution on my YouTube channel - but the lectures will come out some time next year because my schedule is fully booked for this year so you may want to subscribe and click the ‘notifications’ button to be notified when those tutorials are released:

All the best with it.

3

Thanks for your gracious reply. First, I’ll respond about what I want to fix.

I’m a semi-retired pathologist who has a love for H&E microscopy and photomicrography. One of the things I’ve noticed in the past few years is that training for medical students and now even some pathology resident physicians is centered (and tested) primarily on histology images on a screen, particularly whole slide images. I don’t want to be critical of whole slide imaging, but my personal opinion is that while these images may be adequate for diagnosis, they are inferior in terms of aesthetics. I’ve tried to encourage the people I mentor to spend time with a real microscope and learn to appreciate well-stained and well-lit histologic specimens.

To that end, I have distributed some photomicrographs to students and colleagues based on aesthetics. They have generally been well-received (see Pathology cases — Inflammatory myofibroblastic tumor of the humerus – Billoblog )

But here’s my problem. My photomicrographs are “good” but they simply don’t reflect either the dynamic range nor the crispness of what I see through the glass. The images tend to have lower dynamic range (even with histogram stretching) to my eyes and seem just slightly blurred. If you look at the link to my blog, the images are all just a little blurry compared to what I see through glass, and that’s the best I’m getting. I’m not trying to get super-resolution or anything. I want my photomicrographs to be more pleasing. I want them to more accurately reflect what I’m seeing through glass.

My scope is a rather old Olympus BH-2 that I’ve been using for 37 years with a halogen light, but it looks fine through the oculars. I’m assuming that there’s degradation in the lenses in the camera tube. I upgraded cameras from a Nikon D90 to a Z7, with perceivable improvement. I can deal with the chromatic aberration, and spherical aberration is not a big problem (which I try to deal with by simple z stacking if I have to).

My background is in pathology and computer science. I wrote visualization algorithms, including 3D reconstruction stuff for confocal microscopy back in the 1990s) and did software development and system adminstration for the Department of Biophysics and Cellular Pathology at the Armed Forces Institute of Pathology back when it was still around. I am also a forensic pathologist with an interest in patterned injury analysis, particularly the perceptual issues involved in interpreting photographs (see, for instance, https://onlinelibrary.wiley.com/doi/abs/10.1111/1556-4029.13449 ).

I read your 2010 article on generating PSFs from Z stacks. I thought it was great. However, my reading of it gave me the impression that one needed to actually know the amount of z movement in some quantitative way, i.e. use a step motor, not just “turn the fine focus knob a little”. Moreover, my experience is that there are meaningful z stacks only at 40x and 100x, and I’m interested in working on low power images as well.

Thanks, and I enjoyed your articles.

4

I too am a pathologist, retired from clinical work but active in research.
Regarding WSI / digital pathology (DP) the issue here is that clinical pathology research leaders in the field (who earn their living from doing research into digital pathology) have been influential in professional training regulatory circles (The Royal College of Pathologists over here in the UK - and I suspect the CAP in the US) and medical schools. Unfortunately, they seem to have difficulty understanding (or chose to ignore) the value of routine glass slide direct vision microscopy and are calling for this to be scrapped and replaced by WSI DP for all routine diagnosis instead of the more balanced approach of using both side by side. I suspect this is in part because they find it hard to justify the added expense to healthcare managers (who they need to convince to adopt WSI DP in large scale routine healthcare) without having the argument that ‘it will save you money by not having to purchase expensive microscopes for your staff’ - whatever - but the end result is that they strongly advocate the replacement of the microscope with the digital scanner and work that into training programs so young people don’t get a choice and don’t know what they are missing or how they are being de-skilled. This is one reason why I started the PUMA microscope project - to try and help maintain knowledge and skill in direct vision microscopy for those what want it.

Anyways, the issue with WSI DP is that while it may well be a superior method for implementing large scale and routine digital image processing - which can be immensely useful and beneficial - and while it may well be ‘adequate’ for the majority of routine diagnoses, it is undoubtedly an inferior seeing modality for human observers. This is because there has not been (and is unlikely to be in the foreseeable future) any scanner/camera-monitor combo with anywhere close to the same colour and intensity receptivity gamut of the human visual system. This is just a biological and technological fact for anyone who understands both the biology of human vision and the technology of camera-monitor image capture and display. Also nearly all DP WSI high throughput scanners used today do not allow for assessment of birefringence, do not allow variation in opitcs to assess refractility and lack a through-focus control - all of which are easily carried out in seconds on a standard microscope and used frequently by clinical pathologists of my generation to assess inclusion bodies, amyloid and find bugs (like AAFB) as well as assessing whether something is an artefact or a genuine tissue section feature (it is partly due to these deficiencies in DP WSI that even our own RCPath did not recommend DP for diagnostic cytology as opposed to histology (although those things are important for histology too so I disagree with their recommending DP for routine histology). Finally, the WSI scanners used today do not have full Köhler illumination and condenser optics - the highest powers are sometimes ‘interpolated’ from a x20 objective and so on. All these things make DP WSI an inferior seeing modality for clinical diagnostics than using a standard microscope - but, despite all this evidence to the contrary - ‘adeqate’ seems to be the buzz-word in the proponents and they are in the position of decision making power and since no papers are being published that focus on the detrimental effects of DP (it is always about how well DP ‘agrees with’ standard pathology) the effect of these limitations are just not being ‘seen’. Hence the march of DP replacement of microscopes in routine practice will carry on - sadly (for both pathologists and patients).

So it seems that the end result is the replacement of the superior with the adequate for reasons of economy and bias, instead of the better approach which would be to use both side-by-side.

Regarding your problem - It is a very subjective thing. From a scientific point of view you will never be able to get digital images displayed to the same degree of visibility as your direct vision by your eye. But it may be that you can get your images to look good enough for you - however, if that is possible, I suspect the answer is to be found in a selection of digital processing filters, none of which will involve deconvolution.