Posted by
James Pawley on
URL: http://confocal-microscopy-list.275.s1.nabble.com/localization-precision-in-PALM-STORM-tp7219699p7221944.html
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy*****
Hi all,
I have enjoyed the descriptions of how to measure
detector noise contributions in PALM/STORM (and I
am interested to see how the sCMOS cameras
compare with the more common EM-CCD!).
However, while it is certainly important to know
noise levels of the various photodetectors, there
may be other noise sources that may be even more
important.
What about the statistical and other errors
related to the very small numbers of fluorescent
molecules actually localized?
Just because we can localize some fluors, doesn't
mean that those measured are a statistically
random subset of those present. For imaging
setups that use laser light and embedded
specimens, 2/3 of the molecules will be in the
wrong orientation to be significantly excited and
the the chance that the other third will be
excited will vary with their orientation. These
limits probably apply to both the photoactivation
beam and that used for excitation (Might it be
important to have them both polarized in the same
direction?).
In addition, when TIRF is used, excitation also
varies very strongly with the distance of the
fluor from the glass-water or glass-media
interface and the depth of the evanescent field
varies with pol orientation. In some cases, this
interface itself, may affect fluorophor
orientation. In any case, the chance of exciting
a fluor attached to the near side of a structure
only a few hundred nm in size may be
significantly different from the chance of
exciting a fluorophor on its far side. Won't this
distort the structure's apparent shape?
Even if the fluorescent molecule is not prevented
from rotating by steric constraints, may we
assume that its chance of being excited varies
only with its position in the evanescent field?
What about Brownian and other motions? Won't such
motions blurr the image of moving fluorophors
causing them to be discriminated out by software
looking for nice round Airy spots? (Some software
assumes that ovoid blobs must represent two
nearby fluorophors and deletes them,
unintentionally creating a "pile-up error" that
discriminates against areas of high fluorophor
density.) And, in systems where slight
astigmatism has been intentionally introduced to
code for small changes in Z, won't some blobs
that are asymmetric because of motion be
"misplaced"?
At a more extreme level, we are only occasionally
interested in the location or motion of a
fluorescent molecule itself. Usually such a
molecule is used to tag some macromolecule
through a linker or even an antibody. In normal
LM, the optical spatial resolution is so poor
that displacements between fluorophor and
macromolecule are undetectable. Is this true when
the alleged resolution approaches nanometers?
On top of this, in STORM, we have the possibility
that some fluors may be turned on, and off and
then on again. In other words, not all detected
and centroided flashes need represent different
molecules. Should this affect how we view the
results? Has anyone checked for this sort of
double-dipping?
Now I am sure that most of you are already aware
that these factors exist and that they may have
some effect on the final image, but I think that
as PALM/STORM becomes more commonly used, we need
to keep in mind that the unparalleled spatial
resolution of these techniques comes at a price
and that part of this price is that many of the
comforting assumptions that simplify the
interpretation of micrographs made using more
common types of microscopy (Such as: "The
brightness of the image is proportional to the
number of tagged molecules in each location on
the specimen.") may no longer be valid.
Of the above list , I think that the most
important "noise" limitation is the small number
of single-fluorophor "flashes" detected, located
and logged into the image. It is a pity that we
can't display PALM/STORM data not as little blobs
that seem to vary in size and intensity and are
rendered in shades of yellow, orange and black
using the charming "Autumn" LUT, but in such a
way that the color of the displayed pixel makes
evident the number of fluorophors actually
detected within "x" nm of the center of each
pixel (Because of overlap, this number need not
be an integer as it may contain fractional
contributions from many nearby detection events.).
I think that such images might be quite sobering.
Happy 2012
Jim Pawley
***************************************************************************
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 9-21, 2012, UBC, Vancouver Canada
Info:
http://www.3dcourse.ubc.ca/
Application deadline 3/16/2012
"If it ain't diffraction, it must be statistics." Anon. 11/16/12
>To join, leave or search the confocal microscopy listserv, go to:
>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>*****
>
>Hello Christophe,
>As Hendrik has pointed out, there are issues with using the Thompson formula
>and this has been investigated by us and others.
>
>The rigorous approach to calculate the localization accuracy that is based
>on the Fisher information matrix was introduced by our group many years ago
>(e.g, see Ober et al., Biophys. J. 2004, 86:1185-1200, Ram et al., Proc.
>SPIE, 2005, 5699: 426-435, Ram et al., Biophys J. 2008, 95: 6025-6043).
>
>We have a software package to calculate the limit of the localization
>accuracy for different image profiles. Please check out our lab webpage for
>more info (
http://www4.utsouthwestern.edu/wardlab/fandplimittool.asp).
>
>Regards,
>Sripad
>
>
>-----Original Message-----
>From: Confocal Microscopy List [mailto:
[hidden email]] On
>Behalf Of Hendrik Deschout
>Sent: Tuesday, January 24, 2012 11:08 AM
>To:
[hidden email]
>Subject: Re: localization precision in PALM/STORM
>
>*****
>To join, leave or search the confocal microscopy listserv, go to:
>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>*****
>
>Hello Christophe,
>
>If your camera does not show photon counts, you can easily determine the
>conversion factor between the pixel intensity and the number of photons
>yourself, as mentioned by Andrew. The approach exploits the Poisson
>statistics of the photon numbers, see also "Scientific Charge-Coupled
>Devices" by Janesick.
>
>The interpretation of the background term "b" in the Thompson formula is
>somewhat tricky. Someone once pointed out to me that, in case you
>approximate the PSF by a two-dimensional Gaussian, it is the number of
>photons that comes from the background, plus the shoulders of the PSF which
>are not described by the Gaussian. If you again assume a Poisson
>distribution for the photon numbers, "b^2" can be determined as the variance
>of the background pixel intensities near the PSF (corrected with the
>conversion factor of course).
>
>By the way, I would like to note that the Thompson formula (and in
>particular the 30% error) has been corrected by Mortensen et al. in their
>very nice Nature Methods paper (2010, vol. 7, no. 5). Also note that the
>Thompson formula was derived for a CCD camera. In case of an electron
>multiplying CCD camera, the entire expression for "sigma^2" should be
>multiplied by 2, the electron multiplication is to blame, this is often
>ignored. In this case, it also means that the background variance you
>measure is actually equal to "2*b^2".
>
>With kind regards,
>
>Hendrik
>
>Hendrik Deschout
>Biophotonic Imaging Group
>Lab. General Biochemistry and Physical Pharmacy
>Ghent University
>Harelbekestraat 72
>9000 Gent
>Belgium
>Tel: +32 9 264.80.74
>Fax: +32 9 264.81.89
>
[hidden email]
>
>
>-----Original Message-----
>From: Confocal Microscopy List [mailto:
[hidden email]] On
>Behalf Of Andrew York
>Sent: dinsdag 24 januari 2012 16:39
>To:
[hidden email]
>Subject: Re: localization precision in PALM/STORM
>
>*****
>To join, leave or search the confocal microscopy listserv, go to:
>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>*****
>
>Expose your camera to different constant, uniform light levels, perhaps
>with a brightfield lamp. Take a bunch of pictures at each light level. For
>a given pixel, plot variance in signal vs. mean signal at each light level.
>The shape of this curve should tell you how many A/D counts you get per
>photoelectron, since photoelectron counting is hopefully a Poisson process.
>Bonus points: Do all your pixels agree on the relationship between variance
>and mean?
>
>Devil's advocate: suppose you do your calibration wrong. How would you ever
>tell? There ought to be a good answer for this. If there's no way you could
>ever tell you were wrong, what are you accomplishing?
>
>On Tue, Jan 24, 2012 at 9:51 AM, Christian Soeller <
[hidden email]
>> wrote:
>
>> *****
>> To join, leave or search the confocal microscopy listserv, go to:
>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>> *****
>>
>> All cameras effectively run in a photon counting mode (more precisely they
>> count the photo-electrons). The sCMOS cams basically have no EM gain but
>> low enough read-out noise that it is not strictly necessary with typical
>> photon counts. We have briefly tested a Andor Neo and it seems to work.
>>
>> Christian
>>
>> On 25/01/2012, at 3:29 AM, Roger Phillips wrote:
>>
>> > The new scientific CMOS cameras are touted as applicable to single
>> molecule localization microscopy. But I find no mention of a photon
>> counting mode. How are photon counts made with these cameras?
>> > Roger
>> >
>> > Dr Roger Guy Phillips
>> > Centre for Advanced Microscopy,
>> > University of Sussex
>> > School of Life Sciences
>> > John Maynard Smith Building
>> > Falmer, Brighton & Hove
>> > BN1 9QG
>> > United Kingdom
>> >
>> > phone:44 (0)1273 877585
>> > fax: 44 (0)1273 678433
>> > email:
[hidden email]
>> > room:2C9 (ext 7585)/lab 4C2 (ext 2734)
>> >
>> >
>> >
>> >
>> > -----Original Message-----
>> > From: Confocal Microscopy List [mailto:
[hidden email]]
>> On Behalf Of Christophe Leterrier
>> > Sent: 24 January 2012 13:04
>> > To:
[hidden email]
>> > Subject: Re: localization precision in PALM/STORM
>> >
>> > *****
>> > To join, leave or search the confocal microscopy listserv, go to:
>> >
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>> > *****
>> >
>> > Hi Guy,
>> >
>> > Not sure I understand what you mean here, I don't think I'm using
>> "photon count" in my experiment. It's a wide field setup and I'm acquiring
>> streams of EMCCD camera full-frame (512px*512px) images at 10-20 Hz
>> framerate, with a magnified pixel size around 100 nm. The conditions are
>> optimized for acquiring a whole photon burst (500-3000 photons depending
>on
>> the fluorophore) in one, max two images, spread on a few pixels, to have a
>> good signal.
>> >
>> > Do I have to ask Photometrics for a calibration curve between intensity
>> levels and photoelectrons (and relate to photons using the quantum
>> efficiency curve) ? I couldn't find that info on their website.
>> >
>> > Thanks for your help,
>> >
>> > Christophe
>> >
>> >
>> > Le mardi 24 janvier 2012 à 12:06, Guy Cox a écrit :
>> >> Generally one records PALM / STORM images in photon counting mode. That
>> means that signals below a certain threshold are regarded as noise, and
>> discarded, and signals above a higher threshold are regarded as 'pile-up'
>> and also discarded. So there should be no noise level to worry about and
>> each individual image should be classed by your software as a 0 (no
>photon)
>> or a 1 (a photon). The number of counts for that one photon, in a single
>> frame, are meaningless. To estimate the photon counts for a point in the
>> image you need to see in how many frames that point is scored as a 1. I
>> hope this makes sense!
>> >>
>> >> Guy
>> >>
>> >> Optical Imaging Techniques in Cell Biology by Guy Cox CRC Press /
>> >> Taylor & Francis
http://www.guycox.com/optical.htm>> >> ______________________________________________
>> >> Guy Cox, MA, DPhil(Oxon), Honorary Associate, 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 Christophe
>> >> Leterrier
>> >> Sent: Tuesday, 24 January 2012 8:27 PM
>> >> To:
[hidden email]
> > >> (mailto:
[hidden email])
>> >> Subject: localization precision in PALM/STORM
>> >>
>> >> *****
>> >> To join, leave or search the confocal microscopy listserv, go to:
>> >>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>> >> *****
>> >>
>> >> Hi,
>> >>
>> >> Not strictly a confocal question, but I'm pretty sure this list is the
>> best place to get thorough and insightful answers.
>> >>
>> >> I have made 2D STORM (stochastic optical reconstruction microscopy)
>> acquisitions and processing and I end up with a table of XY localized
>> fluorophores together with the integrated intensity of the localized
>> diffraction-limited spot.
>> >>
>> >> I'd like to plot each fluorophore as a gaussian with a width
>> corresponding to the localization precision, similar to what was done in
>> Bates et al. Science 2007. According to equation (17) in Thompson, Larson
>&
>> Webb Biophys J. 2002 (
http://goo.gl/5GIXM), this precision depends on the
>> number of photons collected, the width of the diffraction-limited spot,
>the
>> size of the camera pixel, and the background noise.
>> >>
>> >> So my question is : How do I get the number of photons from the
>> intensity level of an image? I'm using a Photometrics 512*512 QuantEM
>> camera. What is the background noise and how do I estimate it? Then using
>> these values in the Thompson et al. equation, I can get a theoretical spot
>> intensity / localization precision calibration curve that I could use for
>> the gaussian-based reconstruction.
>> >>
>> >> Thanks for your help,
>> >>
>> >> --
>> >> Christophe Leterrier
>> >> Researcher
>> >> Axonal Domains Architecture Team
>> >> CRN2M CNRS UMR 7286
>> >> Aix Marseille University, France
>> >>
>> >>
>> >>
>> >>
>>
>> --
>> Christian Soeller PhD Dept. of Physiology +64 9 3737599 x82770
>> University of Auckland Auckland, New Zealand fax +64 9 3737499
>>
--
***************************************************************************
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 9-21, 2012, UBC, Vancouver Canada
Info:
http://www.3dcourse.ubc.ca/
Application deadline 3/16/2012
"If it ain't diffraction, it must be statistics." Anon. 11/16/12