Resolution and super resolution

[Alberto Diaspro]

*****
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 friends,
here is a contribute to the discussion.

https://www.researchgate.net/publication/322318770_Evaluating_image_resolution_in_stimulated_emission_depletion_microscopy

Best
Alby

[Steffen Dietzel]

*****
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 Alby,

this is a very interesting paper indeed. Fourier ring correlation (FRC)
sounds like a pretty cool approach for determining the resolution in any
sample, for STED or confocal images. What I don't find in the paper or
supplemental is how this approach was realized, like as a Matlab
routine, or a Fiji macro or else?

And then of course there is the question if this is something that could
or will be made available to the public. I have no idea how complicated
these formulas in the supplemental look to a physicist. But as I
biologist I can easily tell that there is no way I could implement this
myself. Thus a respective tool would be very much appreciated. With line
sequential recording one could side-step the drift issue, thus the
actual FRC determination and a guide how to translate the result into
nanometers should be sufficient for widespread application.

Best

Steffen




Am 20.01.2018 um 18:51 schrieb Alberto Diaspro:

> *****
> 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 friends,
> here is a contribute to the discussion.
>
> https://www.researchgate.net/publication/322318770_Evaluating_image_resolution_in_stimulated_emission_depletion_microscopy
>
> Best
> Alby

--
------------------------------------------------------------
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

[Kyle Michael Douglass]

*****
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 Steffen,


On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
>
>
> Thus a respective tool would be very much appreciated.
>
>

Here is a list of a few open-source options for computing the FRC
resolution from image data originating from some different modalities:

1. There is a ImageJ plugin written by the BIOP team at EPFL:
http://imagej.net/Fourier_Ring_Correlation_Plugin

2. Nieuwenhuizen et al. provide MATLAB routines for localization
microscopy data and, if memory serves me correctly, a lightweight ImageJ
plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/

3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
splitting image sequences prior to FRC computation, as well as a
block-wise FRC method:
https://www.biorxiv.org/content/early/2017/07/17/158279

My apologies if I missed any; these are only the ones that I am aware of.

Best regards,
Kyle

--
Kyle M. Douglass, PhD
Post-doctoral researcher
The Laboratory of Experimental Biophysics
EPFL, Lausanne, Switzerland
http://kmdouglass.github.io
http://leb.epfl.ch

[Steffen Dietzel]

*****
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.
*****

Thanks, Kyle,

in particular the first link looks promising. I'll see that I'll get
some twin images asap to test it.

Steffen


Am 22.01.2018 um 16:15 schrieb Kyle Douglass:

> *****
> 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 Steffen,
>
>
> On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
>>
>>
>> Thus a respective tool would be very much appreciated.
>>
>>
>
> Here is a list of a few open-source options for computing the FRC
> resolution from image data originating from some different modalities:
>
> 1. There is a ImageJ plugin written by the BIOP team at EPFL:
> http://imagej.net/Fourier_Ring_Correlation_Plugin
>
> 2. Nieuwenhuizen et al. provide MATLAB routines for localization
> microscopy data and, if memory serves me correctly, a lightweight
> ImageJ plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/
>
> 3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
> splitting image sequences prior to FRC computation, as well as a
> block-wise FRC method:
> https://www.biorxiv.org/content/early/2017/07/17/158279
>
> My apologies if I missed any; these are only the ones that I am aware of.
>
> Best regards,
> Kyle
>

--
------------------------------------------------------------
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

[Ricardo Henriques]

*****
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 Steffen and Kyle,

Kyle, thank you for highlighting our plugin (ref 3 on your list):
https://bitbucket.org/rhenriqueslab/nanoj-squirrel/
<https://bitbucket.org/rhenriqueslab/nanoj-squirrel/wiki/Home>

Steffen, a note that you may enjoy about our method (NanoJ-SQUIRREL), is
that not only does it get you resolution via FRC but also metrics in image
quality. We frequently find that for SR (including STED) neither resolution
nor image quality throughout the field-of-view are homogeneous. SQUIRREL can
give you quantitative perspective into this.

All the best,
-Ricardo

On Tue, 23 Jan 2018 at 16:10 Steffen Dietzel <[hidden email]> 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.
> *****
>
> Thanks, Kyle,
>
> in particular the first link looks promising. I'll see that I'll get
> some twin images asap to test it.
>
> Steffen
>
>
> Am 22.01.2018 um 16:15 schrieb Kyle Douglass:
> > *****
> > 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 Steffen,
> >
> >
> > On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
> >>
> >>
> >> Thus a respective tool would be very much appreciated.
> >>
> >>
> >
> > Here is a list of a few open-source options for computing the FRC
> > resolution from image data originating from some different modalities:
> >
> > 1. There is a ImageJ plugin written by the BIOP team at EPFL:
> > http://imagej.net/Fourier_Ring_Correlation_Plugin
> >
> > 2. Nieuwenhuizen et al. provide MATLAB routines for localization
> > microscopy data and, if memory serves me correctly, a lightweight
> > ImageJ plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/
> >
> > 3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
> > splitting image sequences prior to FRC computation, as well as a
> > block-wise FRC method:
> > https://www.biorxiv.org/content/early/2017/07/17/158279
> >
> > My apologies if I missed any; these are only the ones that I am aware of.
> >
> > Best regards,
> > Kyle
> >
> --
> ------------------------------------------------------------
> 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
>

--
Ricardo Henriques, Associate Professor
<http://www.ucl.ac.uk/lmcb/research-group/ricardo-henriques-research-group>
MRC-Laboratory for Molecular Cell Biology
University College London
Gower Street, London WC1E 6BT, UK
Twitter: @HenriquesLab <https://twitter.com/HenriquesLab>

[Giuseppe Vicidomini]

*****
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 Steffen, Kyle and Riccardo,
happy to see that this topic is gaining your interest.
Our paper want to show the potential of the well-known FRC tool on STED imaging. We was positively surprise about the sensitivity of this tool on many sample-dependent and system-dependent conditions. I really hope that the FRC can become a standard, also tanks to all other excellent paper, packages and plugins already listed here. If you think that it can be useful, I am also happy to share with you our easy-to-use Matlab packages optimised for point scanning imaging. Just send me an e-mail  (giuseppe.vicidomini[at]iit.it)
Best regards
Giuseppe

[Steffen Dietzel]

*****
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 Ricardo,

thank you for your remark. I was under the impression that Squirrel
would need superresolution and 'normal' images for comparison. Upon your
remark I dug deeper into the paper and now assume that it also can be
used to compare any identical images, like two confocal images. Doing it
blockwise is certainly an advantage. So I will test that. And it also is
always good to have independent tools for any given question to see if
they come up with the same result.

My thanks also to Guiseppe for offering the Matlab package. Since I am
familiar with Fiji but not with Matlab, I will try the other two tools
first though. I also want to mention that I did like your paper a lot,
because it nicely explains the basic principle of FRC - which I didn't
quite understand before. And it nicely points out the practical
importance of drift, aka better use line sequential.

Best

Steffen


Am 23.01.2018 um 17:33 schrieb Ricardo Henriques:

> *****
> 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 Steffen and Kyle,
>
> Kyle, thank you for highlighting our plugin (ref 3 on your list):
> https://bitbucket.org/rhenriqueslab/nanoj-squirrel/
> <https://bitbucket.org/rhenriqueslab/nanoj-squirrel/wiki/Home>
>
> Steffen, a note that you may enjoy about our method (NanoJ-SQUIRREL), is
> that not only does it get you resolution via FRC but also metrics in image
> quality. We frequently find that for SR (including STED) neither resolution
> nor image quality throughout the field-of-view are homogeneous. SQUIRREL can
> give you quantitative perspective into this.
>
> All the best,
> -Ricardo
>
> On Tue, 23 Jan 2018 at 16:10 Steffen Dietzel <[hidden email]> 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.
>> *****
>>
>> Thanks, Kyle,
>>
>> in particular the first link looks promising. I'll see that I'll get
>> some twin images asap to test it.
>>
>> Steffen
>>
>>
>> Am 22.01.2018 um 16:15 schrieb Kyle Douglass:
>>> *****
>>> 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 Steffen,
>>>
>>>
>>> On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
>>>>
>>>> Thus a respective tool would be very much appreciated.
>>>>
>>>>
>>> Here is a list of a few open-source options for computing the FRC
>>> resolution from image data originating from some different modalities:
>>>
>>> 1. There is a ImageJ plugin written by the BIOP team at EPFL:
>>> http://imagej.net/Fourier_Ring_Correlation_Plugin
>>>
>>> 2. Nieuwenhuizen et al. provide MATLAB routines for localization
>>> microscopy data and, if memory serves me correctly, a lightweight
>>> ImageJ plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/
>>>
>>> 3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
>>> splitting image sequences prior to FRC computation, as well as a
>>> block-wise FRC method:
>>> https://www.biorxiv.org/content/early/2017/07/17/158279
>>>
>>> My apologies if I missed any; these are only the ones that I am aware of.
>>>
>>> Best regards,
>>> Kyle
>>>
>> --
>> ------------------------------------------------------------
>> 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
>>
>

--
-- ----------------------------------------------------------

Steffen Dietzel, PD Dr. rer. nat.
Head of the Core Facility Bioimaging at the Biomedical Center
Ludwig-Maximilians-Universität München
Walter-Brendel-Zentrum für Experimentelle Medizin

Address:
Biomedical Center
Großhaderner Straße 9
D-82152 Planegg-Martinsried

Phone: [phone]
skype: steffendietzel
e-mail: [hidden email]
fax-to-e-mail: +49/89/2180-9971518
http://www.bioimaging.bmc.med.uni-muenchen.de


--
------------------------------------------------------------
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

[Steffen Dietzel]

*****
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 all,

we now could play a bit with the two Fourier-Ring-Correlation-plugins
for Fiji/ImageJ, the BIOP-FRC and the NanoSquirrel-FRC. Both are easy to
use, differences being that the BIOP-FRC works on two separate images
and squirrel on stacks. A bit annoying with the squirrel is that it does
not suggest the pixel size from the image properties as default, so the
correct pixel size has first to be extracted from the properties an then
entered in the dialog. As Ricardo mentioned, squirrel allows to divide
the image in separate blocks. BIOP-FRC can produce a nice plot for each
comparison which is helpful to understand what is going on. In short,
both have their advantages.

They nicely show that accumulation of photons (lineaccu) improves
resolution, as does closing of the pinhole in confocal images from 1 to
0.5 AU.

What worries me though, is that both plug-ins produce different results
for the same pair of images (squirrel used with 1 block, so the whole
image is looked at). So far, Squirrel always gave larger values (in nm).
Sometimes the difference is small (e.g. 186/190 nm), sometimes it is
rather large (302/372 nm).

So i am wondering why that is. And what it means for practical purposes,
i.e. how reliable is an FRC measurement on real life data. I definitely
see the beauty of using FRC to determine the resolution in an image, but
if the output varies according to unknown reasons, I am not sure I want
to recommend that to our users. But maybe reasons are known? Could it be
noise or SNR?

Thoughts, anybody?

Steffen


Am 25.01.2018 um 13:58 schrieb Steffen Dietzel:

> *****
> 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 Ricardo,
>
> thank you for your remark. I was under the impression that Squirrel
> would need superresolution and 'normal' images for comparison. Upon
> your remark I dug deeper into the paper and now assume that it also
> can be used to compare any identical images, like two confocal images.
> Doing it blockwise is certainly an advantage. So I will test that. And
> it also is always good to have independent tools for any given
> question to see if they come up with the same result.
>
> My thanks also to Guiseppe for offering the Matlab package. Since I am
> familiar with Fiji but not with Matlab, I will try the other two tools
> first though. I also want to mention that I did like your paper a lot,
> because it nicely explains the basic principle of FRC - which I didn't
> quite understand before. And it nicely points out the practical
> importance of drift, aka better use line sequential.
>
> Best
>
> Steffen
>
>
> Am 23.01.2018 um 17:33 schrieb Ricardo Henriques:
>> *****
>> 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 Steffen and Kyle,
>>
>> Kyle, thank you for highlighting our plugin (ref 3 on your list):
>> https://bitbucket.org/rhenriqueslab/nanoj-squirrel/
>> <https://bitbucket.org/rhenriqueslab/nanoj-squirrel/wiki/Home>
>>
>> Steffen, a note that you may enjoy about our method (NanoJ-SQUIRREL), is
>> that not only does it get you resolution via FRC but also metrics in
>> image
>> quality. We frequently find that for SR (including STED) neither
>> resolution
>> nor image quality throughout the field-of-view are homogeneous.
>> SQUIRREL can
>> give you quantitative perspective into this.
>>
>> All the best,
>> -Ricardo
>>
>> On Tue, 23 Jan 2018 at 16:10 Steffen Dietzel <[hidden email]>
>> 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.
>>> *****
>>>
>>> Thanks, Kyle,
>>>
>>> in particular the first link looks promising. I'll see that I'll get
>>> some twin images asap to test it.
>>>
>>> Steffen
>>>
>>>
>>> Am 22.01.2018 um 16:15 schrieb Kyle Douglass:
>>>> *****
>>>> 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 Steffen,
>>>>
>>>>
>>>> On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
>>>>>
>>>>> Thus a respective tool would be very much appreciated.
>>>>>
>>>>>
>>>> Here is a list of a few open-source options for computing the FRC
>>>> resolution from image data originating from some different modalities:
>>>>
>>>> 1. There is a ImageJ plugin written by the BIOP team at EPFL:
>>>> http://imagej.net/Fourier_Ring_Correlation_Plugin
>>>>
>>>> 2. Nieuwenhuizen et al. provide MATLAB routines for localization
>>>> microscopy data and, if memory serves me correctly, a lightweight
>>>> ImageJ plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/
>>>>
>>>> 3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
>>>> splitting image sequences prior to FRC computation, as well as a
>>>> block-wise FRC method:
>>>> https://www.biorxiv.org/content/early/2017/07/17/158279
>>>>
>>>> My apologies if I missed any; these are only the ones that I am
>>>> aware of.
>>>>
>>>> Best regards,
>>>> Kyle
>>>>
>>> --
>>> ------------------------------------------------------------
>>> 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
>>>
>>
>

--
------------------------------------------------------------
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

[Ricardo Henriques]

*****
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 Steffen,

Both BIOP and SQUIRREL use the same engine for FRC (Alex Herbert's
original code).
SQUIRREL does default to the 1/7 threshold while BIOP allows you to both
select different ones + plus gives detailed information on the frequency
profile of the correlation. I think the differences are more about the fact
that SQUIRREL breaks the image into blocks instead of using a pre-defined
ROI.

> So i am wondering why that is. And what it means for practical purposes,
i.e. how reliable is an FRC measurement on real life data.

I would always suggest taking FRC measurements with a grain of salt. There are
many ways the readout can be biased (e.g.: constant patterning/artefacts
across both images). Unfortunately I don't believe we have an unbiased way
to calculate resolution (yet), at the end of the day most of us resort to
looking at line profiles across 2 structures + FRC and use these as a rough
suggestion for resolution.

I would actually welcome the feedback of the community for other
resolution measurement
alternatives.

All the best,
-Ricardo

On Wed, 31 Jan 2018 at 11:45 Steffen Dietzel <[hidden email]> 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.
> *****
>
> Dear all,
>
> we now could play a bit with the two Fourier-Ring-Correlation-plugins
> for Fiji/ImageJ, the BIOP-FRC and the NanoSquirrel-FRC. Both are easy to
> use, differences being that the BIOP-FRC works on two separate images
> and squirrel on stacks. A bit annoying with the squirrel is that it does
> not suggest the pixel size from the image properties as default, so the
> correct pixel size has first to be extracted from the properties an then
> entered in the dialog. As Ricardo mentioned, squirrel allows to divide
> the image in separate blocks. BIOP-FRC can produce a nice plot for each
> comparison which is helpful to understand what is going on. In short,
> both have their advantages.
>
> They nicely show that accumulation of photons (lineaccu) improves
> resolution, as does closing of the pinhole in confocal images from 1 to
> 0.5 AU.
>
> What worries me though, is that both plug-ins produce different results
> for the same pair of images (squirrel used with 1 block, so the whole
> image is looked at). So far, Squirrel always gave larger values (in nm).
> Sometimes the difference is small (e.g. 186/190 nm), sometimes it is
> rather large (302/372 nm).
>
> So i am wondering why that is. And what it means for practical purposes,
> i.e. how reliable is an FRC measurement on real life data. I definitely
> see the beauty of using FRC to determine the resolution in an image, but
> if the output varies according to unknown reasons, I am not sure I want
> to recommend that to our users. But maybe reasons are known? Could it be
> noise or SNR?
>
> Thoughts, anybody?
>
> Steffen
>
>
> Am 25.01.2018 um 13:58 schrieb Steffen Dietzel:
> > *****
> > 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 Ricardo,
> >
> > thank you for your remark. I was under the impression that Squirrel
> > would need superresolution and 'normal' images for comparison. Upon
> > your remark I dug deeper into the paper and now assume that it also
> > can be used to compare any identical images, like two confocal images.
> > Doing it blockwise is certainly an advantage. So I will test that. And
> > it also is always good to have independent tools for any given
> > question to see if they come up with the same result.
> >
> > My thanks also to Guiseppe for offering the Matlab package. Since I am
> > familiar with Fiji but not with Matlab, I will try the other two tools
> > first though. I also want to mention that I did like your paper a lot,
> > because it nicely explains the basic principle of FRC - which I didn't
> > quite understand before. And it nicely points out the practical
> > importance of drift, aka better use line sequential.
> >
> > Best
> >
> > Steffen
> >
> >
> > Am 23.01.2018 um 17:33 schrieb Ricardo Henriques:
> >> *****
> >> 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 Steffen and Kyle,
> >>
> >> Kyle, thank you for highlighting our plugin (ref 3 on your list):
> >> https://bitbucket.org/rhenriqueslab/nanoj-squirrel/
> >> <https://bitbucket.org/rhenriqueslab/nanoj-squirrel/wiki/Home>
> >>
> >> Steffen, a note that you may enjoy about our method (NanoJ-SQUIRREL), is
> >> that not only does it get you resolution via FRC but also metrics in
> >> image
> >> quality. We frequently find that for SR (including STED) neither
> >> resolution
> >> nor image quality throughout the field-of-view are homogeneous.
> >> SQUIRREL can
> >> give you quantitative perspective into this.
> >>
> >> All the best,
> >> -Ricardo
> >>
> >> On Tue, 23 Jan 2018 at 16:10 Steffen Dietzel <[hidden email]>
> >> 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.
> >>> *****
> >>>
> >>> Thanks, Kyle,
> >>>
> >>> in particular the first link looks promising. I'll see that I'll get
> >>> some twin images asap to test it.
> >>>
> >>> Steffen
> >>>
> >>>
> >>> Am 22.01.2018 um 16:15 schrieb Kyle Douglass:
> >>>> *****
> >>>> 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 Steffen,
> >>>>
> >>>>
> >>>> On 01/22/2018 03:24 PM, Steffen Dietzel wrote:
> >>>>>
> >>>>> Thus a respective tool would be very much appreciated.
> >>>>>
> >>>>>
> >>>> Here is a list of a few open-source options for computing the FRC
> >>>> resolution from image data originating from some different modalities:
> >>>>
> >>>> 1. There is a ImageJ plugin written by the BIOP team at EPFL:
> >>>> http://imagej.net/Fourier_Ring_Correlation_Plugin
> >>>>
> >>>> 2. Nieuwenhuizen et al. provide MATLAB routines for localization
> >>>> microscopy data and, if memory serves me correctly, a lightweight
> >>>> ImageJ plugin: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4149789/
> >>>>
> >>>> 3. NanoJ-SQUIRREL by Culley et al. provides ImageJ routines for
> >>>> splitting image sequences prior to FRC computation, as well as a
> >>>> block-wise FRC method:
> >>>> https://www.biorxiv.org/content/early/2017/07/17/158279
> >>>>
> >>>> My apologies if I missed any; these are only the ones that I am
> >>>> aware of.
> >>>>
> >>>> Best regards,
> >>>> Kyle
> >>>>
> >>> --
> >>> ------------------------------------------------------------
> >>> 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
> >>>
> >>
> >
> --
> ------------------------------------------------------------
> 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
>

--
Ricardo Henriques, Associate Professor
<http://www.ucl.ac.uk/lmcb/research-group/ricardo-henriques-research-group>
MRC-Laboratory for Molecular Cell Biology
University College London
Gower Street, London WC1E 6BT, UK
Twitter: @HenriquesLab <https://twitter.com/HenriquesLab>

[Kyle Michael Douglass]

*****
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 Steffen,

Your questions are highly relevant, especially because they are, at
least in my opinion, outstanding and important questions for the meaning
of resolution.

On 01/31/2018 12:43 PM, Steffen Dietzel wrote:
>
>
> What worries me though, is that both plug-ins produce different
> results for the same pair of images (squirrel used with 1 block, so
> the whole image is looked at). So far, Squirrel always gave larger
> values (in nm). Sometimes the difference is small (e.g. 186/190 nm),
> sometimes it is rather large (302/372 nm).
>
> So i am wondering why that is.

Algorithms that compute the FRC are comprised of a series of steps that
are typically performed sequentially. If any one step is implemented
differently between two different packages, then all the following steps
should produce different outputs as well.

One possible difference might be in how the window function is
implemented. In Nieuwenhuizen et al.'s 2013 paper, they implemented a
Tukey window which attenuates spatial frequencies from the edges that
would otherwise contaminate the image's spectrum. Another source of the
difference could be whether the algorithms assume a circularly symmetric
spectrum and compute the correlation on a half ring or full ring. The
BIOP FRC code computes the correlation from a half ring, but they note
that their Fourier transform implementation is not exactly symmetric.
(See line 77 here:
https://c4science.ch/source/ijp-frc/browse/master/src/main/java/ch/epfl/biop/frc/FRC.java 
)

I think it might be best to think of it this way: there is not one FRC.
Rather, there are algorithms that calculate values that we call the FRC.
This is just my opinion after having played with the FRC over the past
few years. I would be happy to hear opposing viewpoints.

> And what it means for practical purposes, i.e. how reliable is an FRC
> measurement on real life data. I definitely see the beauty of using
> FRC to determine the resolution in an image, but if the output varies
> according to unknown reasons, I am not sure I want to recommend that
> to our users. But maybe reasons are known? Could it be noise or SNR?
>

I am most familiar with the use of the FRC in localization microscopy; I
do not know how well what I am about say generalizes to other methods.

In localization microscopy (STORM/PALM/PAINT, etc.) I think it's fairly
well understood that the FRC value is not independent of the sample. A
good explanation for this is found in Supplementary Figure 28 and
Supplemental Note 4 of Legant et al., Nature Methods 13, 359 (2016).
Essentially, the sample's spectrum adds with the spectrum of the label
distribution, so you can get different FRC values from different
structures that have the same labeling density. This can be very
counter-intuitive the first time you use the FRC.

Additionally, the FRC in localization microscopy is biased by multiple
blinking, which can introduce correlations in the spectrum that
otherwise do not reflect the resolution of your measurement.

Given what I know from localization microscopy, I would not be surprised
if there are pitfalls for other modalities. I agree with Ricardo that
you should take the FRC with a grain of salt.

Cheers,
Kyle

--
Kyle M. Douglass, PhD
Post-doctoral researcher
The Laboratory of Experimental Biophysics
EPFL, Lausanne, Switzerland
http://kmdouglass.github.io
http://leb.epfl.ch

[James D. Manton]

*****
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 Ricardo,

> I would actually welcome the feedback of the community for other
> resolution measurement alternatives.
I've always liked the approach of computing the radial average of the
Fourier spectrum of the acquired image and plotting that data in full,
rather than trying to reduce the resolution to a single number. The
'Assessing resolution in super-resolution imaging' paper by Demmerle et
al. (http://dx.doi.org/10.1016/j.ymeth.2015.07.001) has a nice example
of this. This approach nicely reflects that having a system that can
achieve a higher maximum resolution limit is probably not as useful as
one that can achieve a significantly lower maximum resolution but has
much higher transfer strength throughout up to this point.

Best wishes,
James

[Steffen Dietzel]

*****
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 all,

just for completeness, I did use the fixed 1/7 threshold for the
BIOP-FRC and one block only within squirrel. So I expected (near)
identical results when comparing the same image pair.

I can easily imagine that image pairs taken at different areas of the
same sample lead to different results even with the same algorithm and
we indeed found that when looking at beads. Different structures may hit
the edge, or whatever. That's not my point.

But the very same image pair put into different tools with the same
parameters (1 block, threshold 1/7), I had expected more similar
results. When both tools do use the same engine, as you say, Ricardo,
Kyle probably has a point with different pretreatments. I don't read
Java, so I can't check, but it would explain the outcome.


Another thing I still have to wrap my brain around is the fact that for
a 'non-confocal' (open pinhole) image pair of 26 nm beads excited with
645 nm we are getting an FRC resolution of 198/200 nm (BIOP/squirrel,
respectively) while according to 0.5*lambda/NA and lambda=645 nm, NA=1.4
the resolution limit is supposed to be at 230 nm. So it seems we are
getting a resolution that is better than theoretically possible. Not
usually a good sign.


Steffen


Am 31.01.2018 um 13:09 schrieb Ricardo Henriques:

> *****
> 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 Steffen,
>
> Both BIOP and SQUIRREL use the same engine for FRC (Alex Herbert's
> original code).
> SQUIRREL does default to the 1/7 threshold while BIOP allows you to both
> select different ones + plus gives detailed information on the frequency
> profile of the correlation. I think the differences are more about the fact
> that SQUIRREL breaks the image into blocks instead of using a pre-defined
> ROI.
>
>> So i am wondering why that is. And what it means for practical purposes,
> i.e. how reliable is an FRC measurement on real life data.
>
> I would always suggest taking FRC measurements with a grain of salt. There are
> many ways the readout can be biased (e.g.: constant patterning/artefacts
> across both images). Unfortunately I don't believe we have an unbiased way
> to calculate resolution (yet), at the end of the day most of us resort to
> looking at line profiles across 2 structures + FRC and use these as a rough
> suggestion for resolution.
>
> I would actually welcome the feedback of the community for other
> resolution measurement
> alternatives.
>
> All the best,
> -Ricardo
>
>

--
------------------------------------------------------------
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

[Cardone, Giovanni]

*****
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 Steffen,

regarding your concern on the experiment with beads providing a FRC resolution superior to the Abbe resolution limit, I think that such result could be very well plausible if the quality of the two images was high. And this would be because the FRC measure is not really a measure of resolution in the classical definition, rather a measure of quality assessed by verifying the consistency between two images of the same field of view.
Roughly speaking, the FRC is an indirect measure of the SNR in the image, therefore it provides an indication of the size of the features that can be resolved in the image because they are sufficiently above the noise level.
This means that in the case of an experiment with beads, if the background is clean and you use high laser power, you can theoretically reach a FRC resolution up to the Nyquist. This just means that you can resolve the PSF of the instrument very well (e.g. you can see multiple zeros), and not that you have achieved super-resolution.
I think that the FRC can be  a valuable measure to assess the quality of images, if complemented by other measures and if properly interpreted. If you get a value that is worse than the theoretical resolution limit, then you are limited by the noise, while if it is better, then you are limited by the instrument.

Giovanni


-----Original Message-----
From: Confocal Microscopy List [mailto:[hidden email]] On Behalf Of Steffen Dietzel
Sent: Mittwoch, 31. Januar 2018 18:47
To: [hidden email]
Subject: Re: Resolution and super resolution, FRC tools

*****
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 all,

just for completeness, I did use the fixed 1/7 threshold for the BIOP-FRC and one block only within squirrel. So I expected (near) identical results when comparing the same image pair.

I can easily imagine that image pairs taken at different areas of the same sample lead to different results even with the same algorithm and we indeed found that when looking at beads. Different structures may hit the edge, or whatever. That's not my point.

But the very same image pair put into different tools with the same parameters (1 block, threshold 1/7), I had expected more similar results. When both tools do use the same engine, as you say, Ricardo, Kyle probably has a point with different pretreatments. I don't read Java, so I can't check, but it would explain the outcome.


Another thing I still have to wrap my brain around is the fact that for a 'non-confocal' (open pinhole) image pair of 26 nm beads excited with
645 nm we are getting an FRC resolution of 198/200 nm (BIOP/squirrel,
respectively) while according to 0.5*lambda/NA and lambda=645 nm, NA=1.4 the resolution limit is supposed to be at 230 nm. So it seems we are getting a resolution that is better than theoretically possible. Not usually a good sign.


Steffen


Am 31.01.2018 um 13:09 schrieb Ricardo Henriques:

> *****
> 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 Steffen,
>
> Both BIOP and SQUIRREL use the same engine for FRC (Alex Herbert's
> original code).
> SQUIRREL does default to the 1/7 threshold while BIOP allows you to
> both select different ones + plus gives detailed information on the
> frequency profile of the correlation. I think the differences are more
> about the fact that SQUIRREL breaks the image into blocks instead of
> using a pre-defined ROI.
>
>> So i am wondering why that is. And what it means for practical
>> purposes,
> i.e. how reliable is an FRC measurement on real life data.
>
> I would always suggest taking FRC measurements with a grain of salt.
> There are many ways the readout can be biased (e.g.: constant
> patterning/artefacts across both images). Unfortunately I don't
> believe we have an unbiased way to calculate resolution (yet), at the
> end of the day most of us resort to looking at line profiles across 2
> structures + FRC and use these as a rough suggestion for resolution.
>
> I would actually welcome the feedback of the community for other
> resolution measurement alternatives.
>
> All the best,
> -Ricardo
>
>

--
------------------------------------------------------------
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

[Giuseppe Vicidomini]

*****
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 all,
I read your e-mails regarding the application of the FRC in confocal microscopy and I would like to share our experience on that.
Before, let me stress one important aspect of the FRC analysis. The FRC analysis is fully based on the (spatial) frequency content of the image(s). This means that any aspects which influence the frequency content in the images will be reflected in the FRC analysis. Ideally, in a confocal system the images' frequency only depends on the frequency content of the sample (we can assume unlimited) and the characteristics of the microscope (in short the OTF), successively noise “hide” some frequencies in the image.
However, unfortunately, other artefacts can easily enter in the frequencies of a confocal image and affected the FRC analysis. Here a  few common artefacts which apparently improve the frequency band (resolution) above the cut-off frequency (diffraction limit):
- fast diffusion in the sample. FRC analysis share the same framework of image correlation spectroscopy, which is a terrific tool to measure the molecular diffusion properties in the sample. Not surprise, if your FRC analysis shows an increase of the frequency band of the system. In short, intensity fluctuations (temporal frequency) enter in your image as spatial frequency. This problem can be solved by registering the two image non “simultaneously”, e.g. line by line or frame-by-frame in order to cancel out all the temporal correlations. Here, it is interesting to observe that the FRC is a method focused on the concept of static resolution, however more interesting would be to have method to study the dynamic resolution of the system.
- saturation of the detector. This is a problem, that is not always considered. We realise that working in the linear regime of a detector is not always easy, in particular when using pulsed excitation laser and with single-photon detector. Saturation introduces high frequencies in the images, but this frequency content does not belong to the sample. Saturation of the detector is completely different from saturation of the fluorescence, in this case high frequencies belong to the sample. This is the base of many non-linear microscopy technique, such as SAX microscopy.
- any laser fluctuations or vibrations (the drift does not represent a problem). Again, temporal fluctuations enter in the image as spatial frequencies. Collecting the image line by line or frame by frame can cancel this temporal correlation.
Regarding the fact that high SBR and high SNR images produces FRC-based resolution value "above" the diffraction-limits I am not fully convinced. We tested these conditions on synthetic images and the increase of SNR and SBR always provide values in agreement with the diffraction-limit. We did similar experiment on real data increasing the laser power or the pixel-dweel time and the cut-off frequency (diffraction barrier) was overcome only when the detector saturates.
In essence, I am fully agree with you that the FRC is a valuable method to access quality of an image(s) but it must be properly interpreted.
Happy to discuss more on this topic.
Best regards
Giuseppe

> On 8 Feb 2018, at 12:51, Cardone, Giovanni <[hidden email]> 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 Steffen,
>
> regarding your concern on the experiment with beads providing a FRC resolution superior to the Abbe resolution limit, I think that such result could be very well plausible if the quality of the two images was high. And this would be because the FRC measure is not really a measure of resolution in the classical definition, rather a measure of quality assessed by verifying the consistency between two images of the same field of view.
> Roughly speaking, the FRC is an indirect measure of the SNR in the image, therefore it provides an indication of the size of the features that can be resolved in the image because they are sufficiently above the noise level.
> This means that in the case of an experiment with beads, if the background is clean and you use high laser power, you can theoretically reach a FRC resolution up to the Nyquist. This just means that you can resolve the PSF of the instrument very well (e.g. you can see multiple zeros), and not that you have achieved super-resolution.
> I think that the FRC can be  a valuable measure to assess the quality of images, if complemented by other measures and if properly interpreted. If you get a value that is worse than the theoretical resolution limit, then you are limited by the noise, while if it is better, then you are limited by the instrument.
>
> Giovanni
>
>
> -----Original Message-----
> From: Confocal Microscopy List [mailto:[hidden email]] On Behalf Of Steffen Dietzel
> Sent: Mittwoch, 31. Januar 2018 18:47
> To: [hidden email]
> Subject: Re: Resolution and super resolution, FRC tools
>
> *****
> 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 all,
>
> just for completeness, I did use the fixed 1/7 threshold for the BIOP-FRC and one block only within squirrel. So I expected (near) identical results when comparing the same image pair.
>
> I can easily imagine that image pairs taken at different areas of the same sample lead to different results even with the same algorithm and we indeed found that when looking at beads. Different structures may hit the edge, or whatever. That's not my point.
>
> But the very same image pair put into different tools with the same parameters (1 block, threshold 1/7), I had expected more similar results. When both tools do use the same engine, as you say, Ricardo, Kyle probably has a point with different pretreatments. I don't read Java, so I can't check, but it would explain the outcome.
>
>
> Another thing I still have to wrap my brain around is the fact that for a 'non-confocal' (open pinhole) image pair of 26 nm beads excited with
> 645 nm we are getting an FRC resolution of 198/200 nm (BIOP/squirrel,
> respectively) while according to 0.5*lambda/NA and lambda=645 nm, NA=1.4 the resolution limit is supposed to be at 230 nm. So it seems we are getting a resolution that is better than theoretically possible. Not usually a good sign.
>
>
> Steffen
>
>
> Am 31.01.2018 um 13:09 schrieb Ricardo Henriques:
>> *****
>> 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 Steffen,
>>
>> Both BIOP and SQUIRREL use the same engine for FRC (Alex Herbert's
>> original code).
>> SQUIRREL does default to the 1/7 threshold while BIOP allows you to
>> both select different ones + plus gives detailed information on the
>> frequency profile of the correlation. I think the differences are more
>> about the fact that SQUIRREL breaks the image into blocks instead of
>> using a pre-defined ROI.
>>
>>> So i am wondering why that is. And what it means for practical
>>> purposes,
>> i.e. how reliable is an FRC measurement on real life data.
>>
>> I would always suggest taking FRC measurements with a grain of salt.
>> There are many ways the readout can be biased (e.g.: constant
>> patterning/artefacts across both images). Unfortunately I don't
>> believe we have an unbiased way to calculate resolution (yet), at the
>> end of the day most of us resort to looking at line profiles across 2
>> structures + FRC and use these as a rough suggestion for resolution.
>>
>> I would actually welcome the feedback of the community for other
>> resolution measurement alternatives.
>>
>> All the best,
>> -Ricardo
>>
>>
> --
> ------------------------------------------------------------
> 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