Posted by
Guy Cox-2 on
URL: http://confocal-microscopy-list.275.s1.nabble.com/Deconvolution-of-Confocal-Images-was-Airy-Units-tp6947651p6950213.html
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy*****
Jim,
My comment was unnecessarily terse, and I apologise for that. But it does get me cranky when people claim there's some value in the out of focus light. You yourself pointed out that this is crazy, some time ago. Just one micrometre from the focal plane the cone of light from a high NA lens is 6µm in diameter. Go much further and it's everwhere! (This is just repeating what you have said). We can of course work out the intensity distribution expected 1µm from a bright point, and subtract it from that plane (that's deconvolution, or part of it). But unless our sample is trivially sparse we can't predict the bright spot from its ghost at any substantial defocus. (And if our sample is so sparse conventional resolution doesn't apply - that's STORM). So we are not using all detectable photons in any meaningful sense. What is more important, we are not using all excited fluorescence, since each fluorochrome molecule has a finite life.
Confocal images will always be noisier than WF unless we scan for a very long time - we all seem to be agreed on that. But both are exciting fluorescence outside the focal plane. Multiphoton does not - we actually CAN make use of every detected photon. We will need patience, granted, but if people are prepared to take hours acquiring a STORM or PALM dataset maybe they should think about taking several minutes to acquire a multiphoton one? (And then, maybe, deconvolving it).
(I fear that my µ symbols may not make it through the list server, so where you see µm it means micrometres).
Guy
Optical Imaging Techniques in Cell Biology
by Guy Cox CRC Press / Taylor & Francis
http://www.guycox.com/optical.htm______________________________________________
Associate Professor Guy Cox, MA, DPhil(Oxon)
Australian Centre for Microscopy & Microanalysis,
Madsen Building F09, University of Sydney, NSW 2006
Phone +61 2 9351 3176 Fax +61 2 9351 7682
Mobile 0413 281 861
______________________________________________
http://www.guycox.net
-----Original Message-----
From: Confocal Microscopy List [mailto:
[hidden email]] On Behalf Of James Pawley
Sent: Tuesday, 1 November 2011 7:17 AM
To:
[hidden email]
Subject: Re: Deconvolution of Confocal Images? (was: Airy Units)
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy*****
>*****
>To join, leave or search the confocal microscopy listserv, go to:
>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>*****
>
>Sorry Guy to reuse your unnecessarily terse comment, it is you who are wrong.
>
>Multiphoton imaging clearly does limit the
>origins of any fluorescence but does not address
>the original problem, Poisson noise, about why
>confocal images appear noisy - lack of in focus
>photons, which you correctly pointed out.
>Deconvolution of widefield images does address
>Poisson noise since it efficiently uses all
>detectable photons and therefore is worthwhile.
>A deconvolved widefield image will often be
>preferable to a confocal or multiphoton image.
>There is also a semantic question about what
>constitutes noise and if unwanted out of focus
>photons can be returned to their origins then
>they cease to be noise and become signal.
>
Hi again,
Let's recognize that decon is at its base a
mathematical process relating to continuous,
mathematical functions. Here is works perfectly!
Therefore, that although the idea that "inversion
yields microscope truth" has appeal, it has
little to do with Fluorescence Microscopy,
especially that of living cells where the data is
discontinuous (pixellated) and where Poisson
noise is extremely important.
Maybe we need some numbers here. If we assume a
CCD read noise of +/- 5e/pixel, it will be
overshadowed by Poisson noise for any signal
level above 5x5= 25e/pixel. I submit that few
pixels in a widefield CCD data set record a
signal representing less than 25e (not
necessarily 25 counts in the memory but that
number multiplied by the gain-factor of the
camera to represent e/pixel). So Poisson Noise is
always dominant.
On the other hand, Gaussian noise is a lot easier
to model and, if there is a lot of out-of-focus
light (so that the in-plane contrast is say only
1% or even 10% of the total, then most pixels
have the almost the same number of electrons and
the square-root of these various numbers will
vary even less. This is the rationale for
assuming that Poisson Noise to be approximated by
a Gaussian without too much problem in widefield
decon.
As to "which is best":
I start by assuming that all 3D fluorescence
microscope data should be deconvolved.
Therefore, the question comes down to which is
the best method of data collection. Widefield
records some near-focus data that might be useful
and that confocal (may) exclude (depending on the
pinhole size). However, if there are bright
features farther from the focus plane, then they
will produce recorded photons that are probably
less useful. In the end it is all a matter of S/N
and this means statistics.
As a result. the decision depends on the geometry
of the specimen stain distribution and in
particular, the extent to which the greater QE of
the CCD (used in widefield. Greater QE reduces
Poisson Noise on the same light signal.),
compared to that of the PMT (used in confocal.
Depending on wavelength, the PMT is about 10x
worse than the CCD but getting better with the
new photon-counting, GaAsP detectors.), is offset
by the Poisson Noise introduced into the
widefield image by bright, out-of-focus features.
i.e., WF/Decon that does very well on
tissue-culture slides (at the UBC Course, for
instance) may have more problems, with embryos if
(and only if) they contain many stained features
in depth.)
Assuming that the acceptance angle of the
objective is 60 deg, the signal recorded in a
pixel of the size needed to Nyquist sample an
in-focus point object, will be about 80x lower
when the point object is only 1µm out of focus
than when it was in focus. This seems to indicate
that it will be hard to record any useful
(in-focus?) signal from features that are more
than 2 or 3 µm away from the focus plane.
However, in WF, large, bright features much
farther from the focus plane will still produced
significant detectable signal (and its associated
Poisson Noise) and it is hard to see how this can
ever be thought useful.
Confocal and multiphoton will not record any signal from such features.
Personally, I feel that the "put the photons back
where they came from" wording is extremely
misleading.
Yes, we think that the processed 3D image
resembles the original object more closely than
does the raw data, but we seldom have any way to
check this. Essentially, the decon process is no
more immune to error than any other
extrapolation. As the data going into it noisy,
the output also has an error rate and the mere
fact that it looks convincing should not blind us
to the need to remember that its validity rests
on assumptions that may not be valid.
For instance, decon assumes that the PSF is the
same over the whole field of view and at every
successive plane recorded. Neither of these
assumptions are "true" in any general sense (see
Handbook Chapters 11 and 20). An extreme example
is shown in Scanning, 24: 241-246 and also
Chapter 35. Here, the presence of the nucleus in
a living cheek cell not only greatly distorts the
PSF when recording features on its far side, it
also can displace them up to 6 micrometers. I
submit that there is no practical and general
method for "deconvolving" out this sort of
distortion.
In addition, 3D widefield data will include only
part of the light from emitted by features near
the edges (or even just outside) of the data
stack and hence this edge volume cannot be
deconvolved. The larger the PSF used, the larger
this volume of incomplete data will be.
And when time or photosensitivity constraints
limit one to collecting only 2D vs time data, the
confocal/multiphoton approach has much to offer,
So, can I suggest that we be content to be glad
that we have all these nice microscopes without
feeling that we have to win some sort of horse
race?
Also we need to we keep in mind that, as optimal
imaging conditions seldom overlap with optimal
cell viability, we need to remain cautious when
drawing conclusions from the high-resolution
aspects of any of the data that these techniques
produce.
Ciao,
JP
***************************************************************************
Prof. James B. Pawley,
Ph. 608-238-3953
21. N. Prospect Ave. Madison, WI 53726 USA
[hidden email]
3D Microscopy of Living Cells Course, June 10-22, 2012, UBC, Vancouver Canada
Info:
http://www.3dcourse.ubc.ca/ Applications accepted after 11/15/12
"If it ain't diffraction, it must be statistics." Anon.
>Quoting Guy Cox <
[hidden email]>:
>
>>*****
>>To join, leave or search the confocal microscopy listserv, go to:
>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>*****
>>
>>" The photons that are rejected by a pinhole usually come from
>>fluorophores within the specimen - deconvolution can be viewed as an
>>attempt to improve images by putting photons back to where they
>>probably originated, rather than to just reject them. Deconvolution
>>makes a more efficient use of the emitted photons.
>>
>>It is therefore possible to obtain an image by deconvolving a
>>widefield z series that, because of photobleaching and rejection of
>>photons by a pinhole, cannot be obtained from a confocal, even if
>>acquisition time was not the limiting constraint.
>>
>>Only in the narrowest sense, when only a single optical section is
>>required, is Guy correct in regarding out of focus fluorescence as
>>noise - perhaps signal of unwanted origin."
>>
>>WRONG!! Yes, of course the photons come from
>>fluorophores within the specimen. Where else
>>could they come from? But they don't come from
>>where we are looking at and so they cannot be
>>assigned to the plane we are imaging. They
>>belong in a different plane and should be
>>assigned there. And in a confocal stack that
>>is exactly what will happen. So of course
>>there is wasted fluorescence - and if we want
>>to avoid this the answer is rather simple - use
>>2-photon. Then there is NO out of plane
>>fluorescence.
>>
>> Guy
>>
>>
>>Optical Imaging Techniques in Cell Biology
>>by Guy Cox CRC Press / Taylor & Francis
>>
http://www.guycox.com/optical.htm>>______________________________________________
>>Associate Professor Guy Cox, MA, DPhil(Oxon)
>>Australian Centre for Microscopy & Microanalysis,
>>Madsen Building F09, University of Sydney, NSW 2006
>>
>>Phone +61 2 9351 3176 Fax +61 2 9351 7682
>> Mobile 0413 281 861
>>______________________________________________
>>
http://www.guycox.net>>
>>
>>
>>-----Original Message-----
>>From: Confocal Microscopy List
>>[mailto:
[hidden email]] On
>>Behalf Of Jeremy Adler
>>Sent: Monday, 31 October 2011 11:08 PM
>>To:
[hidden email]
>>Subject: Re: Deconvolution of Confocal Images? (was: Airy Units)
>>
>>*****
>>To join, leave or search the confocal microscopy listserv, go to:
>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>*****
>>
>>
>>
>>
>>Quoting Guy Cox <
[hidden email]>:
>>
>>>*****
>>>To join, leave or search the confocal microscopy listserv, go to:
>>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>>*****
>>>
>>>Daniel White wrote:
>>>
>>>" in a confocal you throw away most of the signal, as its out of focus.
>>>So as a result the images are often very noisy. "
>>>
>>>This is often stated but IT IS TOTALLY UNTRUE. What is out of focus is
>>>noise, not signal. If you have no SA (and, honestly, if you are
>>>seriously interested in high-resolution imaging that should be a given)
>>>then a confocal microscope with the pinhole set at 1 Airy diameter
>>>throws away no signal at all. So why are confocal images often noisy?
>>>Well, it's just statistics. If you take a wide-field image with a 1
>>>second exposure each point is exposed for one second. If you take a
>>>confocal image at 512 x 512 for 1 second then each point is exposed for
>>>~4 microseconds. The difference is rather substantial ...
>>>
>>> Guy
>>>
>>>Optical Imaging Techniques in Cell Biology
>>>by Guy Cox CRC Press / Taylor & Francis
>>>
http://www.guycox.com/optical.htm>>>______________________________________________
>>>Associate Professor Guy Cox, MA, DPhil(Oxon)
>>>Australian Centre for Microscopy & Microanalysis,
>>>Madsen Building F09, University of Sydney, NSW 2006
>>>
>>>Phone +61 2 9351 3176 Fax +61 2 9351 7682
>>> Mobile 0413 281 861
>>>______________________________________________
>>>
http://www.guycox.net>>>
>>>
>>>-----Original Message-----
>>>From: Confocal Microscopy List [mailto:
[hidden email]]
>>>On Behalf Of daniel white
>>>Sent: Monday, 31 October 2011 8:30 PM
>>>To:
[hidden email]
>>>Subject: Deconvolution of Confocal Images? (was: Airy Units)
>>>
>>>*****
>>>To join, leave or search the confocal microscopy listserv, go to:
>>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>>*****
>>>
>>>Hi Peter,
>>>
>>>On Oct 31, 2011, at 6:02 AM, CONFOCALMICROSCOPY automatic digest system
>>>wrote:
>>>
>>>>
>>>>Date: Sun, 30 Oct 2011 13:09:10 -0700
>>>>From: Peter Werner <
[hidden email]>
>>>>Subject: Deconvolution of Confocal Images? (was: Airy Units)
>>>>
>>>>An interesting point was made here by Jim Pawley:
>>>>
>>>>>I agree that sampling a bit higher than Nyquist never hurts,
>>>>>especially if you deconvolve (as you always should), but I think
>>>>>that it is a mistake to think that one can "separate" out the noise
>>>>>by decon. I think that noise is pretty fundamental.
>>>>
>>>>I had always heard that if you're doing confocal microscopy, at least
>>>
>>>>point-scanning confocal with a pinhole size of 1AU or smaller, that
>>>>deconvolution was superfluous, because you shouldn't be getting out of
>>>
>>>>focus light. So what is gained by deconvolution when one is sampling
>>>>voxel by voxel?
>>>
>>>in a confocal you throw away most of the signal, as its out of focus.
>>>So as a result the images are often very noisy.
>>>Good contrast.... but high Poisson distributed photon shot noise
>>>from only measuring a handful of photons.
>>>
>>>So usually one needs to do something about that noise...
>>>we want to separate the real signal from the noise.
>>>
>>>Often a Gaussian or mean filter is applied... which suppresses the noise
>>>
>>>by smoothing it out... but it also smooths the real signal, so
>>>effectively you lose
>>>the contrast and resolution that was the whole point of doing confocal.
>>>
>>>The smart way to suppress the noise, but keep the contrast and
>>>resolution
>>>is to do deconvolution.
>>>Deconvolution using a max likelyhood method uses the known shape of the
>>>PSF
>>>to make a best guess model of the real fluorophore distribution in the
>>>sample.
>>>You tell the deconvolution algorithm how noisy the image is (you have to
>>>guess
>>>unless you take 2 images and measure it)
>>>then it attempts to throw out the noise and keep the real signal,
>>>resolution and contrast intact.
>>>
>>>D
>>>
>>>>
>>>>Peter G. Werner
>>>>Merritt College Microscopy Program
>>>
>>>Dr. Daniel James White BSc. (Hons.) PhD
>>>
>>>Leader - Image Processing Facility,
>>>Senior Microscopist,
>>>Light Microscopy Facility.
>>>
>>>Max Planck Institute of Molecular Cell Biology and Genetics
>>>Pfotenhauerstrasse 108
>>>01307 DRESDEN
>>>Germany
>>>
>>>+49 (0)15114966933 (German Mobile)
>>>+49 (0)351 210 2627 (Work phone at MPI-CBG)
>>>+49 (0)351 210 1078 (Fax MPI-CBG LMF)
>>>chalkie666 Skype
>>>
http://www.bioimagexd.net BioImageXD
>>>
http://fiji.sc Fiji - is just ImageJ
>>>(Batteries Included)
>>>
http://www.chalkie.org.uk Dan's Homepages
>>>
https://ifn.mpi-cbg.de Biopolis Dresden Imaging
>>>Platform (BioDIP)
>>>dan (at) chalkie.org.uk
>>>( white (at) mpi-cbg.de )
>>>
>>>-----
>>>No virus found in this message.
>>>Checked by AVG - www.avg.com
>>>Version: 10.0.1411 / Virus Database: 2092/3985 - Release Date: 10/30/11
>>>
>>
>>
>>
>>Jeremy Adler
>>IGP
>>Rudbeckslaboratoriet
>>Daghammersköljdsväg 20
>>751 85 Uppsala
>>Sweden
>>
>>0046 (0)18 471 4607
>>
>>-----
>>No virus found in this message.
>>Checked by AVG - www.avg.com
>>Version: 10.0.1411 / Virus Database: 2092/3985 - Release Date: 10/30/11
>>-----
>>No virus found in this message.
>>Checked by AVG - www.avg.com
>>Version: 2012.0.1834 / Virus Database: 2092/4585 - Release Date: 10/30/11
>>
>
>
>
>Jeremy Adler
>IGP
>Rudbeckslaboratoriet
>Daghammersköljdsväg 20
>751 85 Uppsala
>Sweden
>
>0046 (0)18 471 4607
--
***************************************************************************
Prof. James B. Pawley,
Ph. 608-238-3953
21. N. Prospect Ave. Madison, WI 53726 USA
[hidden email]
3D Microscopy of Living Cells Course, June 10-22, 2012, UBC, Vancouver Canada
Info:
http://www.3dcourse.ubc.ca/ Applications accepted after 11/15/12
"If it ain't diffraction, it must be statistics." Anon.
-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 10.0.1411 / Virus Database: 2092/3985 - Release Date: 10/30/11
-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 2012.0.1834 / Virus Database: 2092/4585 - Release Date: 10/30/11