Posted by
George McNamara on
URL: http://confocal-microscopy-list.275.s1.nabble.com/Deconvolution-advice-tp7579090p7579112.html
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy*****
Hi Jay and Lutz,
I believe Andrew is not going to deconvolve the stripes, instead he is
going to deconvolve the points - see video 1 at
http://code.google.com/p/msim/the ... in the subject line is from:
That's what 'msim' does.
Specifically, we built a new kind of fluorescence microscope, which
we call a 'multifocal structured illumination microscope' (MSIM).
Distinguishing features:
* 3D superresolution (145 nm transverse, 400 nm axial resolution)
* Uses standard fluorescent probes (like GFP), currently
configured for 488 and 561 nm illumination
* Works in thick samples (>50 microns). Previous SIM scopes
couldn't do this.
* Reasonable speed (1 2D slice per second)
* Live-cell compatible (tens or hundreds of 3D volumes)
* Low photobleaching (similar to a spinning disk confocal)
* 50x50 micron field of view
Functionally, it's a lot like a point-scanning confocal with twice
the resolution and higher signal-to-noise, at the cost of digital
post-processing.
His Nature Methods paper
http://www.nature.com/nmeth/journal/v9/n7/full/nmeth.2025.html mentions
that the 3D superresolution values he measured, 145 nm XY, 300 nm Z, was
with a standard objective lens - not hand picked:
"The apparent full-width at half maximum (FWHM) intensity of
microtubules in MSIM images was 145 nm, a twofold improvement
compared to wide-field imaging (Fig. 1d). Similar experiments on
110-nm beads confirmed this result (MSIM FWHM of 146 nm ± 15 nm
versus wide-field FWHM of 284 nm ± 32 nm (± s.d.), n = 80 beads,
Supplementary Fig. 7). The acquisition time for the 48 ?m × 49 ?m
field (Fig. 1a) was ~1 s, more than a 6,500-fold improvement over
ISM, assuming the same 222-Hz raw frame rate."
and,
Like existing 3D SIM16, MSIM provides resolution doubling by
increasing the highest spatial frequencies encoded in the raw data,
followed by deconvolution. However, the absolute resolutions we
report are slightly lower than existing 3D SIM, which we attribute
to two factors. First, achieving the highest possible resolution
relies on obtaining diffraction-limited wide-field performance
before (M)SIM is applied. Although we attempted this, the wide-field
point spread functions (PSFs) we measured (Supplementary Fig. 7) had
FWHM (~280 nm) larger than theoretical predictions. It is customary
in SIM to screen many objectives and use only the one with highest
resolution. We did not do this. Furthermore, we suspect that
aberration caused by warping of the dichroic in our microscope was
the major cause of departure from ideal performance, as we noticed
that small adjustments of the mounting screws that held the dichroic
drastically altered the PSF. A thicker, stiffer dichroic may
mitigate this issue. Second, the excitation patterns in previous SIM
implementations concentrate energy at the very highest lateral
spatial frequencies in the objective pass-band. In contrast, the
multifocal excitation patterns we used contain all spatial
frequencies permitted by the objective. We might thus expect
line-excitation--based SIM to provide a greater SNR at high spatial
frequencies than MSIM, enabling higher resolution. We note the
point-like nature of our excitation pattern offers a route to even
higher resolution, perhaps by exploiting the nonlinearities provided
by reversible, switchable fluorescent probes29.
Compared to previous SIM implementations, our system is simpler to
build. We add only a DMD, a telescope and a fast camera to a
conventional wide-field microscope. Our illumination path has no
moving parts and is insensitive to polarization. Finally, we do not
modify the illumination path for multicolor operation, unlike
existing SIM implementations30.
Upshot: MSIM is likely to get even better!
George
On 10/2/2012 12:40 PM, Lutz Schaefer wrote:
> *****
> To join, leave or search the confocal microscopy listserv, go to:
>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy> *****
>
> Hello all,
>
> Jay is correct, a regularized inverse filter of the separated
> (shifted) orders can be used followed by magnitude (and/or linear)
> reconstruction. To the best of my knowledge, this principle is
> implemented in the Zeiss ZEN and AxioVision software. It is also
> possible to use a Gauss likelihood iterative algorithm, but to no
> advantage as the extracted orders cannot be constrained to positivity.
> A realization of Poisson likelihood will, as Jay says proof difficult,
> owing to its inherent nonlinearity.
>
> Best Regards
> Lutz
>
> __________________________________
> L u t z S c h a e f e r
> Sen. Scientist
> Mathematical modeling / Image processing
> Advanced Imaging Methodology Consultation
> 16-715 Doon Village Rd.
> Kitchener, ON, N2P 2A2, Canada
> Phone/Fax: +1 519 894 8870
> Email:
[hidden email]
> ___________________________________
>
> --------------------------------------------------
> From: "Unruh, Jay" <
[hidden email]>
> Sent: Tuesday, October 02, 2012 10:17
> To: <
[hidden email]>
> Subject: Re: Deconvolution advice
>
>> *****
>> To join, leave or search the confocal microscopy listserv, go to:
>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>> *****
>>
>> Hi all,
>>
>> Deconvolution for SIM is a very different story than for other
>> techniques. SIM by default uses an inverse filter in its
>> reconstruction to recombine the shifted components of the fourier
>> transform without enhancing the high frequency noise. Typical SIM
>> software has a noise parameter for the wiener filter. If you set
>> this filter low, you start to see a rippling pattern in the noise and
>> eventually in the high signal regions. As far as I can tell, no one
>> has tried to use more advanced (poisson noise driven) algorithms for
>> this problem.
>>
>> If one assumes that Autoquant is using some variant of the algorithm
>> shown in Tim Holmes' Handbook of Biological Confocal chapter, then
>> this algorithm is not immediately applicable to the SIM problem. The
>> algorithm update function involves dividing the original image by the
>> convolved object guess and then convolving that ratio with the
>> reflected psf and finally multiplying by the object guess. Given
>> that the raw SIM image is actually a frequency modulated image in a
>> particular direction, it is not clear how the original ratio would be
>> generated. Richardson-Lucy has a similar problem. Would you
>> deconvolve before reconstructing? This is the only circumstance
>> under which the noise could be considered poisson. In that case, can
>> the original PSF be used? I'm not entirely certain that the
>> deconvolution would preserve the frequency modulation in the image.
>> In addition, the original reconstruction algorithm would require a
>> second step of deconvolution during the reconstruction step--not sure
>> how the noise parameter should be chosen after initial deconvolution.
>>
>> Jay
>>
>> -----Original Message-----
>> From: Confocal Microscopy List
>> [mailto:
[hidden email]] On Behalf Of Gitta Hamel
>> Sent: Monday, October 01, 2012 8:42 AM
>> To:
[hidden email]
>> Subject: Re: Deconvolution advice
>>
>> *****
>> To join, leave or search the confocal microscopy listserv, go to:
>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>> *****
>>
>> **commercial response**
>>
>>
>> Hello Andrew,
>>
>> It's fully understandable that people want to know the scientific
>> grounds when using Huygens.
>>
>> For the full list of articles I refer to
>>
http://www.svi.nl/HuygensReferences at which the relevant papers are
>> at the bottom of the page and mostly written during the years 1996-1998.
>> There are much more articles that ought to be included so your
>> question shows that we must give more attention to this topic.
>>
>> With best wishes,
>>
>> Gitta Hamel
>>
>> ****************************************
>> Gitta Hamel
>> Managing Director Scientific Volume Imaging bv Developers of the
>> *HUYGENS* software The Netherlands
>> phone: ++ 31 35 6 42 16 26
>> *****************************************
>>
>>
>> ^SVI Customer support: mail us your questions
[hidden email]
>> <mailto:
[hidden email]>or find answers online in our Huygens
>> WIKI:www.svi.nl/FrontPage <
http://%20www.svi.nl/FrontPage>
>>
>>
>>
>> On 09/29/2012 04:00 PM, Gens, John Scott wrote:
>>> *****
>>> To join, leave or search the confocal microscopy listserv, go to:
>>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>> *****
>>>
>>> Andrew-
>>>
>>> You might want to get in touch with Jim McNally. Last I heard he was
>>> at NIH-NCI.
>>>
>>> Some of his older papers on deconvoltion algorithms are below, but he
>>> can probably point you towards more recent information.
>>>
>>>
http://www.ncbi.nlm.nih.gov/pubmed/10579932>>>
>>>
http://www.ncbi.nlm.nih.gov/pubmed/11541650 ( in particular, fig.2
>>> compared a 3D image processed by three different algorithms)
>>>
>>>
>>> Quoting Andrew York <
[hidden email]>:
>>>
>>>> *****
>>>> To join, leave or search the confocal microscopy listserv, go to:
>>>>
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy>>>> *****
>>>>
>>>> Hello, I'm looking for advice and information about deconvolution,
>>>> especially from those with first-hand experience.
>>>>
>>>> Traditionally, one of the processing steps in structured illumination
>>>> microscopy is deconvolution. For our SIM, we decided to use an
>>>> open-source
>>>> solution:
>>>>
https://sites.google.com/site/piotrwendykier/software/deconvolution/p>>>> aralleliterativedeconvolution
>>>>
>>>>
>>>> This seemed like a nice tradeoff between reinventing the wheel with
>>>> our own deconvolution code, and subjecting ourselves to a 'black box'
>>>> closed-source
>>>> solution. However, we've recently tried out the Huygens deconvolution
>>>> software, and the results seem quite promising, possibly an
>>>> improvement over other methods we've tried. I like good images, but I
>>>> don't like black boxes, and I like to understand my data processing.
>>>>
>>>> 1. Is the exact algorithm used in Huygens transparently documented
>>>> anywhere? I spent a few hours searching today, but if it's out there,
>>>> I missed it.
>>>>
>>>> 2. Is there a clear winner for deconvolution algorithms? What should
>>>> I be using?
>>>>
>>>> 3. Are there other deconvolution software packages I should consider?
>>>> Ideally I'm looking for software based on clearly-documented
>>>> algorithms.
>>>>
>>>> Thanks for the help.
>>>>
>>>> -Andrew York
>>>> NIH/NIBIB
>>>>
>>>
>>
>> --
>> Managing Director
>>
>> Huygens SVI
>>
>> tel: +31 (0)35 642 16 26
>>
>> fax: +31 (0)35 683 79 71
>>
>> skype: gittahamel
>>
>> cell: +31(0)618 021272
>>
>> Visiting address
>>
>> Laapersveld 63,
>> 1213 VB Hilversum,
>> The Netherlands
>