The Basic Principle of Airyscanning

[Andrew York]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

 I just read "The Basic Principle of Airyscanning":
http://www.zeiss.com/microscopy/en_de/products/confocal-microscopes/landingpages-airyscanning/airyscanning-technology-note-download.html
 and I greatly enjoyed it. I think Zeiss has made a very cool instrument
here, and I think they've done an excellent job communicating the essence
of how/why it works.

 The paper contains a few errors. I think the errors are worth correcting,
because otherwise the paper is excellent. Comments below:

Page 5:
"""The intensities after pixel reassignment are summed up before
deconvolution so this information about light contribution from
out-of-focus is lost. Therefore the further gain in resolution is limited
to the lateral plane. The axial direction stays merely confocal ...  In
addition, they also have lost the axial information and therefore obtain
resolution enhancement only in the lateral direction.""" (similar
statements on page 6)

This isn't true, but it's interesting why. I discuss this in an earlier
email:
http://lists.umn.edu/cgi-bin/wa?A2=ind1407&L=confocalmicroscopy&T=0&F=&S=&P=19025

Page 5:
"""Methods using cameras are slow because the rate limiting step will be
the readout time of this detector."""

They compare the 880 to instant SIM, and conclude camera-based methods are
slow. I'm not sure where this conclusion comes from. Instant SIM is
camera-based, and its at least 10x faster than the 880. (The only reason we
didn't publish >100 fps is we didn't have interesting samples moving that
fast). Rescan confocal and OPRA are currently slower than the 880, but
that's nothing to do with the cameras, I think it's just because they
didn't make their scan mirrors go fast.

Limiting factors for imaging speed are a fun topic: I expect the 880, like
any single-point-scanning technique, to be primarily limited by excitation
saturation (A very small volume is glowing at any instant, and you can't
get GFP past 100% excited, so there's only so much light per second you can
get). The next limiting factor is data flow rate: how many gigabytes per
second can your detector produce? sCMOS cameras spit out >1 GB/s; I don't
know how fast the 880's detector spits out data, but Airyscanning requires
30 measurements per voxel, compared to 1 measurent per voxel for
hardware-based techniques, so it would have to give >30 GB/s before imaging
framerates could possibly compare. I bet it doesn't give 30 GB/s. Even
then, a computer that can handle 30 GB/s is an exceptional computer.

Other, less important details:

Page 4:
"""As confocal images are noisier [than widefield]..."""

Generally true only in optically thin samples, of course; false in densely
tagged samples with lots of out-of-focus fluorophores.

Page 4:
"""As the overlap decreases with larger displacements, the width of the
resulting Airy disk gets even slightly narrower concomitant with smaller
amplitudes... what is captured with a displaced pinhole contains therefore
a higher proportion of higher frequencies ... the displaced pinhole images
contain disproportionally higher amounts of better localized emitters
compared to the image of the non-displaced pinhole. The larger the
displacement, the higher the proportion of better localized emitters."""
(other similar statements on page 5)

This might not be true at all depending on the precise shape of your PSF;
it's a small effect at most. For example, if your emission and excitation
PSFs resemble a Gaussian, resolution is totally independent of pinhole
displacement. Since PSF shape is invariably aberrated to some degree, take
this with a grain of salt.

Page 5:
"""Although these hardware solutions are instant without the need of image
processing they lack flexibility as data cannot be manipulated in different
ways after acquisition."""

I discussed digital/analog tradeoffs here:
http://lists.umn.edu/cgi-bin/wa?A2=ind1407&L=confocalmicroscopy&T=0&F=&S=&P=20324
Briefly, I agree, postprocessing is nice, but our MSIM users never
want/need/use it. This also ignores the primary drawbacks of digital
processing: more read noise and big data files.

Page 8:

"""A 20x / NA 0.8 objective in Airyscan will yield a performance comparable
to a 63x / NA 1.4 objective employed in a conventional confocal."""

Maybe the lateral resolution is comparable, but I bet the axial resolution
and light collection efficiency are worse.

[Zdenek Svindrych]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

I agree that the paper is very well written. I especially appreciate it's
well balanced and unbiased, not an add!
I think the errors listed by Andrew are not so serious and most of them are
well explained in the paper: the PSF sharpening with pinhole offset is given
just minor importance and it is shown in Eqs (3) and (5) of box 4 (page 16)
that Gaussian approximation does not show this effect. The z-resolution of
traditional ISM is more important issue, but as I remember it was shown in
some papers that it is the same as regular confocal. Well, the 'Virtual NA
effect' sounds commercial and I would prefer to avoid it.
As to errors, i seems that some minus (-) signs are missing in Eq (5) of box
4, but maybe it's just typesetting problem.
One key thing to allways remember is that if you collect 100 photons in a
pixel, you inevitably get +/-10 photons error. Subtracting two such pixel
gives you 0 +/- 14 photons but blocking the light physically gives you
simply zero (the readout noise of modern cameras is < 1 photon). And the
paper fairly states that regular SIM gives better contrast/superresolution
in thin/sparse samples...
The only thing I'm really missing is sample pictures. They should be able to
get some by now :-).
Cheers, zdenek svindrych


---------- Původní zpráva ----------
Od: Andrew York <[hidden email]>
Komu: [hidden email]
Datum: 16. 8. 2014 0:56:22
Předmět: The Basic Principle of Airyscanning

"*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

I just read "The Basic Principle of Airyscanning":
http://www.zeiss.com/microscopy/en_de/products/confocal-microscopes/
landingpages-airyscanning/airyscanning-technology-note-download.html
and I greatly enjoyed it. I think Zeiss has made a very cool instrument
here, and I think they've done an excellent job communicating the essence
of how/why it works.

The paper contains a few errors. I think the errors are worth correcting,
because otherwise the paper is excellent. Comments below:

Page 5:
"""The intensities after pixel reassignment are summed up before
deconvolution so this information about light contribution from
out-of-focus is lost. Therefore the further gain in resolution is limited
to the lateral plane. The axial direction stays merely confocal ... In
addition, they also have lost the axial information and therefore obtain
resolution enhancement only in the lateral direction.""" (similar
statements on page 6)

This isn't true, but it's interesting why. I discuss this in an earlier
email:
http://lists.umn.edu/cgi-bin/wa?A2=ind1407&L=confocalmicroscopy&T=0&F=&S=&P=
19025

Page 5:
"""Methods using cameras are slow because the rate limiting step will be
the readout time of this detector."""

They compare the 880 to instant SIM, and conclude camera-based methods are
slow. I'm not sure where this conclusion comes from. Instant SIM is
camera-based, and its at least 10x faster than the 880. (The only reason we
didn't publish >100 fps is we didn't have interesting samples moving that
fast). Rescan confocal and OPRA are currently slower than the 880, but
that's nothing to do with the cameras, I think it's just because they
didn't make their scan mirrors go fast.

Limiting factors for imaging speed are a fun topic: I expect the 880, like
any single-point-scanning technique, to be primarily limited by excitation
saturation (A very small volume is glowing at any instant, and you can't
get GFP past 100% excited, so there's only so much light per second you can
get). The next limiting factor is data flow rate: how many gigabytes per
second can your detector produce? sCMOS cameras spit out >1 GB/s; I don't
know how fast the 880's detector spits out data, but Airyscanning requires
30 measurements per voxel, compared to 1 measurent per voxel for
hardware-based techniques, so it would have to give >30 GB/s before imaging
framerates could possibly compare. I bet it doesn't give 30 GB/s. Even
then, a computer that can handle 30 GB/s is an exceptional computer.

Other, less important details:

Page 4:
"""As confocal images are noisier [than widefield]..."""

Generally true only in optically thin samples, of course; false in densely
tagged samples with lots of out-of-focus fluorophores.

Page 4:
"""As the overlap decreases with larger displacements, the width of the
resulting Airy disk gets even slightly narrower concomitant with smaller
amplitudes... what is captured with a displaced pinhole contains therefore
a higher proportion of higher frequencies ... the displaced pinhole images
contain disproportionally higher amounts of better localized emitters
compared to the image of the non-displaced pinhole. The larger the
displacement, the higher the proportion of better localized emitters."""
(other similar statements on page 5)

This might not be true at all depending on the precise shape of your PSF;
it's a small effect at most. For example, if your emission and excitation
PSFs resemble a Gaussian, resolution is totally independent of pinhole
displacement. Since PSF shape is invariably aberrated to some degree, take
this with a grain of salt.

Page 5:
"""Although these hardware solutions are instant without the need of image
processing they lack flexibility as data cannot be manipulated in different
ways after acquisition."""

I discussed digital/analog tradeoffs here:
http://lists.umn.edu/cgi-bin/wa?A2=ind1407&L=confocalmicroscopy&T=0&F=&S=&P=
20324
Briefly, I agree, postprocessing is nice, but our MSIM users never
want/need/use it. This also ignores the primary drawbacks of digital
processing: more read noise and big data files.

Page 8:

"""A 20x / NA 0.8 objective in Airyscan will yield a performance comparable
to a 63x / NA 1.4 objective employed in a conventional confocal."""

Maybe the lateral resolution is comparable, but I bet the axial resolution
and light collection efficiency are worse."

[Andrew York]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

>
> The z-resolution of traditional ISM is more important issue, but as I
> remember it was shown in some papers that it is the same as regular
> confocal.


 Probably from De Luca and Manders' excellent rescan confocal paper. They
calculate axial sectioning though, not axial resolution. I agree that axial
sectioning is the same in both cases, but axial resolution is generally
different. I want to emphasize this distinction, before it gets accepted as
conventional wisdom that "hardware SIM doesn't help axial res", like Zeiss
says.

One key thing to always remember is that if you collect 100 photons in a
> pixel, you inevitably get +/-10 photons error. Subtracting two such pixel
> gives you 0 +/- 14 photons but blocking the light physically gives you
> simply zero (the readout noise of modern cameras is < 1 photon).


That reminds me - does anybody know the readout noise/QE of the airyscan
detector? I'm guessing 30 doses of read noise per voxel equals about ~6x
more effective read noise than a single-pixel detector? If the read noise
is <<1e-, this is totally ignorable. If it's more than 1 e-, then 6 to 12
e- of extra noise per voxel is substantial.


>

The only thing I'm really missing is sample pictures.
>

 There're some pretty nice images in their product brochure:
http://www.zeiss.com/microscopy/en_us/products/confocal-microscopes/lsm-880.html#downloads