*****
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.
*****

Dear Colleagues:
Whenever this debate comes up I like to remind that the biggest mistake
people seem to make when trying to use deconvolution in confocal microscopy
is that they have the spatial sampling way too large. In the case of a
confocal microscope with a 1.4 NA objective, the proper sampling is about
50 nm in X,Y and 150 nm in Z. The matter is explained in this excellent
paper by Rainer Heintzmann:

https://www.researchgate.net/publication/47867069_Band_Limit_and_Appropriate_Sampling_in_Microscopy

The reason for this is that the limit (in terms of wavelength) of the
frequency space support for a widefield fluorescence microscope is
Lambda/2NA, whereas for a confocal it is Lambda/4NA. Therefore the sampling
for confocal has to be Lambda/8NA, or about 500/(8*1.4) = 45 nm.

Also i would like to remind that deconvolution is a technique in digital
image processing that is now decades old, and these methods do actually
work. However there are some important "tips and tricks," one being the
re-scaling of intensities in the result. This is usually done because the
deconvolved image can have values that go over the original range (probably
16 bit). If this happens, the image might not be displayed correctly
(especially in programs like photoshop), with over-range values either
being truncated or wrapped around, possibly giving some really strange
looking results.

PS - about photon counting detectors: Traditionally the limit for photon
counting is about 5MHz because of pile-up, in which photon signals start to
bunch together and are no longer counted one at a time. If we are scanning
at 1 image per second, this means that you can only get 5 million photons
per image. A typical confocal microscope image might be much brighter than
this, and so older photon counting methods may not work correctly. Maybe
now there are photon counting setups that can handle much larger count
rates (I guess FLIM is sort of its own story with the 'histogramming'
approach used by B&H).

best,
Guy Hagen




On Thu, May 17, 2018 at 8:22 AM, <[log in to unmask]> wrote:

> *****
> 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.
> *****
>
> Hi James,
> the way I couple detectors to the objective for non-descanned 2P imaging
> is
> by relaying the BFP of the objective lens to the active area of the
> detector. I do it because (1) you couple (almost) all the light that the
> objective lens is able to collect, and (2) the illumination of the
> detector
> is quite homogeneous (the sample itself is perfectly blurred on the
> detector).
>
> With your lens (40x / 1.2 NA) as an example (assuming Oil, Zeiss) with BFP
> diameter around 10 mm, and Hamamatsu hybrid detector with 3 mm diameter
> active area, a simple telecentric relay (f1 = 156 mm, f2 = 50 mm) will do
> the trick. The only question is, how big the two lenses should be? This of
> course depends on the required field of view, and it turns out to be
> (paraxial approcxiamtion, 22 mm field number) 32 mm and 25 mm diameters
> of
> the two relay lenses. Now accounting for the defoucus of several tens of
> µm
> due to light scattering, the relay lenses need to be bigger, but not
> dramatically. For defocus of +/-1% of the objective focal length the
> lenses
> grow to 36 mm and 29 mm diameter. Possibly less with more advanced design.
>
>
> Of course to couple ALL the light that can possibly pass through the
> objective, even bigger relay lenses, or different approach, would be
> needed,
> but that depends on the particularities of each individual objective
> lens...
>
> Best, zdenek
> --
> Zdenek Svindrych, Ph.D.
> Research Associate - Imaging Specialist
> Department of Biochemistry and Cell Biology
> Geisel School of Medicine at Dartmouth
> email: [log in to unmask]
>
> ---------- Původní e-mail ----------
> Od: JAMES B PAWLEY <[log in to unmask]>
> Komu: [log in to unmask]
> Datum: 17. 5. 2018 0:42:27
> Předmět: Re: confocal detectors and deconvolution
> "*****
> 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.
> *****
>
> Hi Craig,
>
> Your analysis is correct BUT only for light that proceeds in straight
> lines
> from the in-focus plane.
>
> Non-rescanned detection is usually used with 2P microscopy to detect light
> that was excited at a particular time (rather than location, as this has
> been defined by the 2P excitation itself). Using time alone to decode
> location allows one to utilize signal photons that have been scattered
> before reaching the objective. This works optimally if you place a PMT
> that
> is totally insensitive to the 2P excitation directly under the specimen.
> Few
> systems manage this and the fallback is to collect light that enters the
> objective. You don’t need much optical analysis to show that light
> emerging
> from a scattering event a few tens of µm towards the objective will emerge
> from this lens as a rapidly diverging ray bundle that doesn’t in any way
> concentrate at the BFP. Any optical components (apart from a light pipe)
> used to try to capture any such rays will only make the situation worse.
>
> Even large SiPM sensors are only 6 mm on a side, considerably smaller than
> the BFP of a 40x NA.1.2NA objective. Consequently, as installing a "beam-
> splitter photodetector" immediately behind the objective is difficult,
> PMTs
> with photocathodes many mm in diameter work best for this sort of methode-
> non-descanned detection because they can intercept more of this rapidly-
> diverging beam.
>
> Best,
>
> Jim P.
> ****************************************
> James and Christine Pawley, 5446 Burley Place (PO Box 2348), Sechelt, BC,
> Canada, V0N3A0,
> Phone 604-885-0840, email <[log in to unmask]>
> NEW! NEW! AND DIFFERENT Cell (when I remember to turn it on!)
> 1-604-989-6146
>
>
> > On May 15, 18, at 11:06 PM, Craig Brideau <[log in to unmask]>
> wrote:
>
> >
> > *****
> > 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.
> > *****
> >
> > When designing an optical system for 2P (or confocal for that matter)
> the
> > scanned beam has to be pivoted around a virtual point located at the
> back
> > aperture of the objective. If this plane was relayed to the face of the
> > large area detector you would effectively have a descanned detection
> path,
>
> > loosely speaking.
> >
> > Craig
> >
> > On Tue, May 15, 2018 at 7:25 PM Michael Giacomelli <[log in to unmask]>
> wrote:
> >
> >> *****
> >> 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.
>
> >> *****
> >>
> >> Hi James,
> >>
> >> That is a very interesting post. I've read about SiPMs before, but
> >> did not realize they were so far as long. The need to evenly
> >> illuminate a large area is slightly annoying (probably not going to
> >> work well for non-descanned 2P), but could be designed around for a
> >> descanned system as you suggest. The specs certainly look
> >> interesting, especially the decreased sensitivity to strong light, and
> >> the very low multiplicative noise.
> >>
> >> I wonder if anyone has tried one in a scanning confocal or 2p system
> >> and compared to conventional PMTs (e.g. h7422s or similar)? I did a
> >> quick search, but didn't see many papers using them for conventional
> >> confocal scanning.
> >>
> >> Mike
> >>
> >> On Mon, May 14, 2018 at 10:50 PM, JAMES B PAWLEY <[log in to unmask]>
> >> wrote:
> >>> *****
> >>> 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.
> >>> *****
> >>>
> >>> Hi Zdenek,
> >>>
> >>> Nice slide show reference! Thanks. I will steal some images!
> >>>
> >>> However…
> >>>
> >>> I think it is a shame that they do not really discuss the difference
> in
> >> “excess noise” between PMTs and SiPMs as this can affect effective QE
> very
> >> strongly.
> >>>
> >>> Also, I think that they could have done a bit more to separate
> >> applications involving diffuse light from those involving
> almost-coherent
>
> >> light, like scanning microscopy, in which the signal originates from a
> very
> >> small volume (and therefore can be focussed onto any surface that is
> even
>
> >> slightly bigger). Of course, PMTs can have huge photocathodes (the R877-
> 100
> >> is 12,860 mm2) and, depending on their temp and IR sensitivity, can
> have
> >> very low dark count rates per square mm of photocathode. But in
> scanning
> >> light microscopy, there seems to be little need for many square mm of
> >> active area.
> >>>
> >>> And even PMTs can have dark current problems. The first version of the
> >> Zeiss 510 used a large number (23? Don’t remember exactly.) of tiny
> >> servo-motors in which the windings were always excited. This heated the
> >> inside of the optics box enough to produce so many dark counts from
> their
>
> >> Hamamatsu PMT modules that they had to replace the servos with DC
> versions
> >> that were only excited when you changed some setting. And this was long
> >> before GaAsP photocathodes were common. From slide 19, you can see that
> the
> >> dark count rate of GaAsP is about 100x that of the more common
> bi-alkali
> >> PMTs.
> >>>
> >>> All the same, silicon diodes do leak current at room temperature and
> so
> >> it is not surprising that most MPPCs (or SiPMs) are supplied with
> Peltier
>
> >> cooling systems.
> >>>
> >>> For most low-light, confocal applications it would seem that the
> S14420
> >> series might work. It has a 1.5 x 1.5 mm array containing 900, 50µm
> cells
>
> >> with a fill factor of 81%, a PDE of 40% and a maximum dark count of
> 1,000
>
> >> cps (or only 0.01 count per (long!) 10µs pixel).
> >>>
> >>> https://www.hamamatsu.com/resources/pdf/ssd/s14420_
> series_kapd1061e.pdf
> >>>
> >>> I know that a max PDE of 40% might sound less than the 40% QE on the
> >> plots of GaAsP PMT photocathodes, but these are different specs: PMT QE
> >> only refers to the fraction of photons making photoelectrons. About 30%
> of
> >> these photoelectrons fail to propagate down the electron-multiplier
> chain
>
> >> (i.e., they are lost to the signal). If you also include the fact that
> the
> >> MPPC has virtually no excess noise, this makes at least an additional
> 40%
>
> >> improvement in the effective QE.
> >>>
> >>> BUT you have to make sure that the ray bundle fills the array.
> >>>
> >>> (OK, maybe you don’t. Let’s assume that 600 of the 900 APDs on this
> >> device fall within a circle that just touches the sides of the square
> and
>
> >> that your signal ray-bundle just fills this circle (uniformly!) when
> your
>
> >> pinhole is open all the way. If we guess (from Slide 22 of the slide
> show)
> >> that the effective RC decay time is 40ns and choose a pixel time of 1
> µs
> >> (or 25 decay times), a simple-minded analysis would lead us to conclude
> >> that a (massive) signal of 600 +/-24 detected photons/1 µs-pixel would
> lose
> >> close to 25 photon pulses due to pulse pile-up (i.e., a number similar
> the
> >> Poisson Noise uncertainty). A detected signal of 64 +/- 8 photons/1µs
> pixel
> >> from a darker area would lose less than one count, an error much
> smaller
> >> than the Poisson noise.
> >>>
> >>> However, if you keep the same signal levels but use a pinhole that is
> >> 3.3 x smaller in diameter, the signal will now hit only 60 APDs and a
> 600
>
> >> PE signal will lose 40% because of pulse pile-up. On the other hand, if
> the
> >> signal getting through the smaller pinhole is has gone down with the
> >> pinhole area and is now only 64 +/-8 pulses, then the loss is only 4%,
> or
>
> >> about 3 counts; smaller than the Poisson noise. For 10µs pixels, the
> signal
> >> would go up by 10 times but the percentage loss would be the same..
> >>>
> >>> So using this detector, you would have another thing to worry about:
> “At
>
> >> this signal level and this pinhole size, am I losing signal to pileup?”
> >>>
> >>> Annoying, but a geometric factor that the computer could easily keep
> >> track of and warn you when there was a problem.
> >>>
> >>> A worse problem is that, if the bundle diameter is larger than the
> >> back-projected image of the Airy Disk, the light in the bundle is NOT
> >> evenly distributed As long as there isn’t much spherical aberration,
> the
> >> image of the spot at the pinhole is always a LOT brighter in the
> centre.
> So
> >> this might be a potential problem, one that could be reduced by using a
> >> larger array (3x3 or 6x6 mm) and changing the optics to fill these
> arrays.
> >> Larger arrays cost more and would increase dark counts by about 4x or
> 16x
>
> >> (resp.) but probably still not enough to worry about.
> >>>
> >>> Happy designing.
> >>>
> >>> Jim Pawley
> >>>
> >>> ****************************************
> >>> James and Christine Pawley, 5446 Burley Place (PO Box 2348), Sechelt,
> >> BC, Canada, V0N3A0,
> >>> Phone 604-885-0840, email <[log in to unmask]<mailto:[log in to unmask]>>
>
> >>> NEW! NEW! AND DIFFERENT Cell (when I remember to turn it on!)
> >> 1-604-989-6146
> >>>
> >>> On May 14, 18, at 9:23 AM, [log in to unmask]<mailto:[log in to unmask]>
>
> >> wrote:
> >>>
> >>> *****
> >>> 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<http://www.imgur.com/> and
> include
> >> the link in your posting.
> >>> *****
> >>>
> >>>
> >>> To chill down the excitement around MCCPs (or SiPMs, as Hamamtsu call
> >> them),
> >>> note that their dark noise is some 3 - 4 orders of magnitude higher
> than
>
> >>> that of regular PMTs! Of course, chilling the detector to -80 degC
> >> (which is
> >>> common with EMCCDs, for example) would solve this issue, but the cost
> >> would
> >>> be prohibitive...
> >>> For more details on SiPMs vs PMTs see here:
> >>>
> >>> https://drive.google.com/open?id=1R3hDR0KX0nE5qWh5GrzyELzBoiGdP1MB
> >>>
> >>>
> >>> Cheers, zdenek
> >>>
> >>>
> >>>
> >>> ---------- Původní e-mail ----------
> >>> Od: JAMES B PAWLEY <[log in to unmask]<mailto:[log in to unmask]>>
> >>> Komu: [log in to unmask]<mailto:
> >> [log in to unmask]>
> >>> Datum: 14. 5. 2018 0:27:35
> >>> Předmět: Re: confocal detectors and deconvolution
> >>> "*****
> >>> 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<http://www.imgur.com/> and
> include
> >> the link in your posting.
> >>> *****
> >>>
> >>> Hi all,
> >>>
> >>> I would like to Eecho Michael’s points.
> >>>
> >>> Because hybrid photodetectors have a very high gain (>>1,000x) in
> their
> >>> first stage, they produce very little “excess noise” (also called
> >>> Multiplicative Noise). As a result, it is possible to characterize
> their
>
> >>> output as "25 photons detected” (although it might be safer to think
> of
> >> it
> >>> as "25 photons detected this time" or ”25 +/- 5 photons.").
> >>>
> >>> Straight PMTs do not share this feature and because single
> photoelectrons
> >>> produce output pulses that vary significantly in size, even the very
> best
> >>> PMTs produce an uncertainty in the magnitude of the signal presented
> to
> >> the
> >>> ADC that is at least 40% larger in relative terms than would be the
> case
>
> >> in
> >>> the absence of this excess noise. On PMTs having electron multipliers
> >>> optimized for other reasons (such as making them very small, like
> those
> >> in
> >>> the 32-PMT linear arrays), the increase in uncertainty is closer to
> 100%
>
> >> (i.
> >>> e., The signal has the same uncertainty that it would have if 4 times
> >> fewer
> >>> photons were counted perfectly.)
> >>>
> >>> Either type of PMT can have a GaAsP photocathode but it will need to
> be
> >>> cooled.
> >>>
> >>> Although single APDs may have a high photon detection efficiency (PDE,
> a
>
> >>> spec that is like QE but which includes the signal lost by
> photoelectrons
> >>> that do not avalanche at all) they have such massive excess noise that
> >> it is
> >>> essential to use them with pulse-counting circuits and these circuits
> are
> >>> just too slow for use in beam-scanning light microscopy.
> >>>
> >>> The solution is the multi-pixel photon counter (MPPC, a development
> from
>
> >> the
> >>> SPAD (single photon avalanche device),
> >>>
> >>> https://www.hamamatsu.com/resources/pdf/ssd/mppc_kapd0004e.pdf
> >>>
> >>>
> >> https://www.hamamatsu.com/us/en/community/optical_sensors/
> articles/sipm_
> the_
> >>> ultimate_photosensor/index.html ).
> >>>
> >>> The surface of an MPPC is covered with an array of 400 to 20,000 APDs,
> >> each
> >>> connected to the high-voltage rail through its own damping resister.
> The
>
> >>> resistor causes the voltage across the APD to drop as the avalanche
> >>> proceeds. This quenches the discharge and produces single-photon
> pulses
> >> of
> >>> very uniform size. As all the APDs are electrically in parallel, these
> >>> single-photon current pulses simply add up producing an output current
> >>> signal almost devoid of excess noise. What could be better? And in
> >> addition
> >>> they are significantly less sensitive than hybrid PMTs to overheating
> >> damage
> >>> if accidentally exposed to a bright light.
> >>>
> >>> There are of course limitations: 1) A significant fraction (20-40%) of
> >> the
> >>> MPPC's surface is taken up with the resistors and the wiring to
> provide
> >> each
> >>> APD with + and - voltages. Photons absorbed or reflected in these
> areas
> >> are
> >>> lost. 2) The system is only free of pulse-pileup losses to the extent
> >> that
> >>> no APD absorbs more than one photon within its RC relaxation time (set
> by
> >>> the R of the resistor, and the capacitance (C) of the sensitive area
> of
> >> the
> >>> APD. Larger individual APDs “waste” proportionally fewer photons
> hitting
>
> >> the
> >>> resistor and wiring, (increasing their effective QE) but this
> increases
> >>> their capacitance (making them more susceptible to pulse-pileup).
> >>>
> >>> All will be well as long as the number of photons absorbed in the
> active
>
> >>> areas during time period RC is small (5%?) compared to the number of
> >> APDs in
> >>> the array THAT ARE ILLUMINATED BY THE BEAM.
> >>>
> >>> The caps above are to remind everyone that, to work properly, the size
> of
> >>> the ray bundle striking the MPPC must be matched to the size of the
> APD
> >>> array (1.3 to 6 mm square). This can be a problem if we imagine the
> ray
> >>> bundle being limited by a confocal aperture that can be varied in size
> >> over
> >>> a substantial range. (Do we need a zoom lens to make all possible
> signal
>
> >> ray
> >>> -bundles match the size of the MPPC array?)
> >>>
> >>> Apart from this, I would like to second the comment that deep imaging
> is
>
> >>> usually limited by spherical aberration and systems that can correct
> for
>
> >>> this without “bumping the specimen while you try to adjust the collar”
> >> are
> >>> to be preferred.
> >>>
> >>> I would also like to reaffirm that, assuming the pixel size meets
> >> Nyquist,
> >>> you should ONLY evaluate results after deconvolving the data with an
> >>> appropriate 2D or 3D PSF. Although the smallest real object in a
> Nyquist
> -
> >>> sampled image will be at least 4 (more likely 5) pixels wide, all the
> >> noise
> >>> terms affect single pixel values. i.e., they have frequency components
> at
> >>> least 4x higher than that which can represent any real structure. In
> >> scanned
> >>> fluorescent imaging, one deconvolves more to reduce noise than to
> >> increase “
> >>> spatial resolution” (although you can also increase resolution as long
> as
> >>> you have massive amounts of signal.)
> >>>
> >>> Cheers,
> >>>
> >>> Jim Pawley
> >>> ****************************************
> >>> James and Christine Pawley, 5446 Burley Place (PO Box 2348), Sechelt,
> BC,
> >>> Canada, V0N3A0,
> >>> Phone 604-885-0840, email <[log in to unmask]<mailto:[log in to unmask]
> >>> <mailto:[log in to unmask]>>
> >>> NEW! NEW! AND DIFFERENT Cell (when I remember to turn it on!)
> >> 1-604-989-6146
> >>>
> >>>
> >>> On May 11, 18, at 10:16 AM, MODEL, MICHAEL <[log in to unmask]<mailto:
> >> [log in to unmask]><mailto:mmodel@
> >>> KENT.EDU<http://KENT.EDU>>> wrote:
> >>>
> >>> *****
> >>> 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.
> >>> *****
> >>>
> >>> We have potential users who want to quantify some kind of small
> >> aggregates
> >>> in the brain. I am afraid that deconvolution can make noise look like
> >> such
> >>> aggregates. Perhaps collecting a noisy image twice and comparing two
> >>> deconvolved images might help, but that seems too much work. Am I
> wrong?
>
> >>>
> >>> -----Original Message-----
> >>> From: Confocal Microscopy List <[log in to unmask]
> <mailto:
>
> >> [log in to unmask]>> On Behalf
> >>> Of Steffen Dietzel
> >>> Sent: Friday, May 11, 2018 10:49 AM
> >>> To: [log in to unmask]<mailto:
> >> [log in to unmask]>
> >>> Subject: Re: confocal detectors and deconvolution
> >>>
> >>> *****
> >>> 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.
> >>> *****
> >>>
> >>> Am 10.05.2018 um 17:19 schrieb Vitaly Boyko:
> >>> there is no big difference between HyDs and GaAsP detectors.
> >>>
> >>> I disagree on this. In my view, the major difference is that the HyD
> >> always
> >>> operates in photon counting mode whether, as far as I know, the PMTs
> >> (with
> >>> or without GaAsP) create an electron cloud of which the size is
> >> determined
> >>> by the number of photoelectrons AND statistics, and the cloud size is
> >> then
> >>> digitized. So the output created by one photon may vary substantially
> >>> depending on the number of electrons created on the first dynodes
> (which
>
> >> in
> >>> turn is a statistical process). My information may be outdated and
> newer
>
> >>> PMTs might have extra tricks, if so please correct me.
> >>>
> >>> Another difference is apparently the size of the photcathode. If
> memory
> >>> serves me right, the larger cathode of the GaAsP PMTs (compared to
> HyDs)
>
> >>> creates more dark noise. I like our HyDs a lot, I appreciate having a
> >> gray
> >>> value of "21 photons" instead of some random number. But having said
> >> this,
> >>> at the end of the day what counts is the sensitivity of the whole
> system,
> >>> and not of the detector alone. So to do this right there is no
> substitute
> >>> for testing your own samples on different machines with your
> >> applications in
> >>> mind.
> >>>
> >>> As for deconvolution, yes, it can create artefacts. But so does
> confocal
>
> >>> microscopy (a point becomes an Airy pattern, not a point). And if you
> do
>
> >> it
> >>> right the deconvolved image will be closer to the truth than the
> original
> >>> image. Should you have the third edition of the handbook around, have
> a
> >> look
> >>> at the preface, last paragraph.
> >>>
> >>> Steffen
> >>>
> >>> --
> >>> ------------------------------------------------------------
> >>> Steffen Dietzel, PD Dr. rer. nat
> >>> Ludwig-Maximilians-Universität München
> >>> Biomedical Center (BMC)
> >>> Head of the Core Facility Bioimaging
> >>>
> >>> Großhaderner Straße 9
> >>> D-82152 Planegg-Martinsried
> >>> Germany
> >>>
> >>> http://www.bioimaging.bmc.med.uni-muenchen.de
> >>>
> >>> "
> >>>
> >>
>
> "
>