Simulation of a fluorescence lifetime image with fixed lifetime in one pixel

[Aleem]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear All,

I have very interesting situation, I want to generate (simulate) a
fluorescence lifetime image of 64X64 pixels  with following
characteristics:
1. Containing 64 peak channels for each pixel
2. Center pixel having 3 ns lifetime
3. Rest of the pixels with zero counts

Is there any way to do this simulation? If yes, could you please suggest
which software should I use ?

Thank you very much for your help.

Aleem

--
*Best Regards,
*
*Aleem Syed
*
*Graduate Assistant
*
*Emily Smith Research Group
*
*0712 Gillman Hall
*
*Department of Chemistry
*
*Iowa State University
*
*Ames Laboratory
*
*IA, USA
*
*Work Ph. # 515-294-8586
*
*
  *

[Unruh, Jay]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Aleem,

Do you want all pixels blank or with noise?  What do you mean by 64 peak channels?  I have an ImageJ plugin that will generate a 64 x 64 pixel FLIM image with a Gaussian instrument response function and whatever multiexponential lifetime you choose.  You can download it from my website (http://research.stowers.org/imagejplugins) along with the source code.  It is under Plugins>Sim Tools>sim FLIM jru v1.  You would simply have to set all pixels other than the center to zero.  In ImageJ, if you set your minimum displayed intensity to zero, set your background color to black, select the center pixel, and run Edit>Clear Outside, it will do what you want.

Jay

-----Original Message-----
From: Confocal Microscopy List [mailto:[hidden email]] On Behalf Of Aleem Syed
Sent: Thursday, April 25, 2013 2:38 PM
To: [hidden email]
Subject: Simulation of a fluorescence lifetime image with fixed lifetime in one pixel

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear All,

I have very interesting situation, I want to generate (simulate) a fluorescence lifetime image of 64X64 pixels  with following
characteristics:
1. Containing 64 peak channels for each pixel 2. Center pixel having 3 ns lifetime 3. Rest of the pixels with zero counts

Is there any way to do this simulation? If yes, could you please suggest which software should I use ?

Thank you very much for your help.

Aleem

--
*Best Regards,
*
*Aleem Syed
*
*Graduate Assistant
*
*Emily Smith Research Group
*
*0712 Gillman Hall
*
*Department of Chemistry
*
*Iowa State University
*
*Ames Laboratory
*
*IA, USA
*
*Work Ph. # 515-294-8586
*
*
  *

[Aleem]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Jay,

Thank you for your reply. I really appreciate it. Initially, I was thinking
to have blank pixels, but now I want to add noise and also I want save this
FLIM data as .sdt or .asc file, so that I can export it to SPCImage.

Any suggestions?

Thanks again.

Aleem




On Thu, Apr 25, 2013 at 4:28 PM, Unruh, Jay <[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Aleem,
>
> Do you want all pixels blank or with noise?  What do you mean by 64 peak
> channels?  I have an ImageJ plugin that will generate a 64 x 64 pixel FLIM
> image with a Gaussian instrument response function and whatever
> multiexponential lifetime you choose.  You can download it from my website (
> http://research.stowers.org/imagejplugins) along with the source code.
>  It is under Plugins>Sim Tools>sim FLIM jru v1.  You would simply have to
> set all pixels other than the center to zero.  In ImageJ, if you set your
> minimum displayed intensity to zero, set your background color to black,
> select the center pixel, and run Edit>Clear Outside, it will do what you
> want.
>
> Jay
>
> -----Original Message-----
> From: Confocal Microscopy List [mailto:[hidden email]]
> On Behalf Of Aleem Syed
> Sent: Thursday, April 25, 2013 2:38 PM
> To: [hidden email]
> Subject: Simulation of a fluorescence lifetime image with fixed lifetime
> in one pixel
>
> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Dear All,
>
> I have very interesting situation, I want to generate (simulate) a
> fluorescence lifetime image of 64X64 pixels  with following
> characteristics:
> 1. Containing 64 peak channels for each pixel 2. Center pixel having 3 ns
> lifetime 3. Rest of the pixels with zero counts
>
> Is there any way to do this simulation? If yes, could you please suggest
> which software should I use ?
>
> Thank you very much for your help.
>
> Aleem
>
> --
> *Best Regards,
> *
> *Aleem Syed
> *
> *Graduate Assistant
> *
> *Emily Smith Research Group
> *
> *0712 Gillman Hall
> *
> *Department of Chemistry
> *
> *Iowa State University
> *
> *Ames Laboratory
> *
> *IA, USA
> *
> *Work Ph. # 515-294-8586
> *
> *
>   *
>

*  *

[Alessandro Esposito]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear Aleem,
    a lot depends on the level of accuracy with which you wish to generate
synthetic data. I mention the best way you could use, although it may be far
too complex for the specific test you wish to run, simply to give you an
overview of the methodologies you could apply.

You could run a Markov Chain Monte Carlo (MCMC) where you can simulate
electronic states of the fluorophore(s), the excitation and the detection
process. These will be described only with transition probabilities between
different molecular states or probability of detection, etc. The advantage of
MCMC is that you will obtain data containing the right amount of noise and
proper decays with the least possible assumptions.

This strategy is very useful when you wish to test theory or analysis
algorithms, but over-kill in most applications. The simplest alternative is to just
generate an exponential decay convoluted with an IRF normalize this to 1 and
then multiply to the number of photons you wish to simulate. After that, you
should add Poissonian noise and you will have a realistic decay. TCSPC is
dominated by Poissonian noise and, therefore, it is not an issue neglecting
other potential sources of noise.

Exporting the data to SPCImage is certainly possible. I have to say, however,
that my experience with creating software to import data from SPCM was a bit
frustrating even with some help from the community. I spent several weeks to
write software  import SDT and SPC files. Perhaps it was just me. I do not
have software to write SDT files though. The good news is that with the
installation of SPCM you have already code in C that should help you in the
process. Also, last time I spoke with B&H, I was told they may have some
code to give for this. I believe for reading, but perhaps also for writing. Also,
you may wish to read the B&H handbook where the SDT file format is
described. My issues with it were only related to differences between different
languages in defining words.

The last bit on which language to use. For a fast project of a good start, any
programming language you are proficient with is good enough. I have also
noticed Jay's reply and you can certainly benefit from his code in Java.
Personally, I develop software in Matlab, very efficient to do full MCMCs or
simplified versions of simulations. Matlab is very popular in the imaging
community and it comparatevly simple to self learn thanks to excellent
documentation. However, unless you have an academic license around... it is a
bit expensive (worth it s money though). Java and C/C++ will do the job,
Octave may be a good (free) substitute for Matlab and R is re-emerging as a
good tool for programming.

If you wish to test the impact of empty bins on lifetime estimation or the
influence of B&H "binning" on datafit, perhaps I can add a few comments on
that.

Kind regards,

Alessandro
www.quantitative-microscopy.org/


On Thu, 25 Apr 2013 13:38:22 -0600, Aleem Syed <[hidden email]>
wrote:

>*****
>To join, leave or search the confocal microscopy listserv, go to:
>http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
>*****
>
>Dear All,
>
>I have very interesting situation, I want to generate (simulate) a
>fluorescence lifetime image of 64X64 pixels  with following
>characteristics:
>1. Containing 64 peak channels for each pixel
>2. Center pixel having 3 ns lifetime
>3. Rest of the pixels with zero counts
>
>Is there any way to do this simulation? If yes, could you please suggest
>which software should I use ?
>
>Thank you very much for your help.
>
>Aleem
>
>--
>*Best Regards,
>*
>*Aleem Syed
>*
>*Graduate Assistant
>*
>*Emily Smith Research Group
>*
>*0712 Gillman Hall
>*
>*Department of Chemistry
>*
>*Iowa State University
>*
>*Ames Laboratory
>*
>*IA, USA
>*
>*Work Ph. # 515-294-8586
>*
>*
>  *

[Aleem]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Alessandro,


Thank you for your reply, I would be glad to have your comments on the
"binning effect" in B&H's SPCImage. I think its not that straightforward as
it sounds. I agree with you on having a code from B&H for at least reading,
to understand what is going on in the software, not just trusting the
output from software blindly. I've tried searching for source code but
failed to find one.

Thanks,

Aleem


On Fri, Apr 26, 2013 at 12:45 AM, Alessandro Esposito <
[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Dear Aleem,
>     a lot depends on the level of accuracy with which you wish to generate
> synthetic data. I mention the best way you could use, although it may be
> far
> too complex for the specific test you wish to run, simply to give you an
> overview of the methodologies you could apply.
>
> You could run a Markov Chain Monte Carlo (MCMC) where you can simulate
> electronic states of the fluorophore(s), the excitation and the detection
> process. These will be described only with transition probabilities between
> different molecular states or probability of detection, etc. The advantage
> of
> MCMC is that you will obtain data containing the right amount of noise and
> proper decays with the least possible assumptions.
>
> This strategy is very useful when you wish to test theory or analysis
> algorithms, but over-kill in most applications. The simplest alternative
> is to just
> generate an exponential decay convoluted with an IRF normalize this to 1
> and
> then multiply to the number of photons you wish to simulate. After that,
> you
> should add Poissonian noise and you will have a realistic decay. TCSPC is
> dominated by Poissonian noise and, therefore, it is not an issue neglecting
> other potential sources of noise.
>
> Exporting the data to SPCImage is certainly possible. I have to say,
> however,
> that my experience with creating software to import data from SPCM was a
> bit
> frustrating even with some help from the community. I spent several weeks
> to
> write software  import SDT and SPC files. Perhaps it was just me. I do not
> have software to write SDT files though. The good news is that with the
> installation of SPCM you have already code in C that should help you in the
> process. Also, last time I spoke with B&H, I was told they may have some
> code to give for this. I believe for reading, but perhaps also for
> writing. Also,
> you may wish to read the B&H handbook where the SDT file format is
> described. My issues with it were only related to differences between
> different
> languages in defining words.
>
> The last bit on which language to use. For a fast project of a good start,
> any
> programming language you are proficient with is good enough. I have also
> noticed Jay's reply and you can certainly benefit from his code in Java.
> Personally, I develop software in Matlab, very efficient to do full MCMCs
> or
> simplified versions of simulations. Matlab is very popular in the imaging
> community and it comparatevly simple to self learn thanks to excellent
> documentation. However, unless you have an academic license around... it
> is a
> bit expensive (worth it s money though). Java and C/C++ will do the job,
> Octave may be a good (free) substitute for Matlab and R is re-emerging as a
> good tool for programming.
>
> If you wish to test the impact of empty bins on lifetime estimation or the
> influence of B&H "binning" on datafit, perhaps I can add a few comments on
> that.
>
> Kind regards,
>
> Alessandro
> www.quantitative-microscopy.org/
>
>
> On Thu, 25 Apr 2013 13:38:22 -0600, Aleem Syed <[hidden email]>
> wrote:
>
> >*****
> >To join, leave or search the confocal microscopy listserv, go to:
> >http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> >*****
> >
> >Dear All,
> >
> >I have very interesting situation, I want to generate (simulate) a
> >fluorescence lifetime image of 64X64 pixels  with following
> >characteristics:
> >1. Containing 64 peak channels for each pixel
> >2. Center pixel having 3 ns lifetime
> >3. Rest of the pixels with zero counts
> >
> >Is there any way to do this simulation? If yes, could you please suggest
> >which software should I use ?
> >
> >Thank you very much for your help.
> >
> >Aleem
> >
> >--
> >*Best Regards,
> >*
> >*Aleem Syed
> >*
> >*Graduate Assistant
> >*
> >*Emily Smith Research Group
> >*
> >*0712 Gillman Hall
> >*
> >*Department of Chemistry
> >*
> >*Iowa State University
> >*
> >*Ames Laboratory
> >*
> >*IA, USA
> >*
> >*Work Ph. # 515-294-8586
> >*
> >*
> >  *
>

--
*Best Regards,
*
*Aleem Syed
*
*Graduate Assistant
*
*Emily Smith Research Group
*
*0712 Gillman Hall
*
*Department of Chemistry
*
*Iowa State University
*
*Ames Laboratory
*
*IA, USA
*
*Work Ph. # 515-294-8586
*
*Cell Ph. # 515-708-8283
  *

[Alessandro Esposito]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear Aleem,
    perhaps it would help if you specify in more detail which is the issue you are
trying to explain.

However, a general note of caution when using the "binning" option in
SPCImage is necessary.

First of all it is not technically binning. It is a convolution averaging filter. It is
slower than binning, but a good idea as it permits to improve SNR with a better
compromise in respect of loss in spatial resolution resolution.

Second, the bin value you input in the GUI is not the dimension of the kernel.
The kernel dimension is 2*x+1.

Therefore, who does not read the manual, with x=2 may think to bin 2x2 pixel
together, but actually this would be an averging filter with a kernel of 5x5
pixel.

Third, the default option of visualization in SPCImage is to overlay lifetime
maps with intensity maps. Beautiful representation that provide the
advantages of a smoother lifetime image (when using "binning") with the
perception of higher resolution provided by the overlay with the non-binned
intensity image.

All legit and described by B&H. However, users should always inspect the "raw"
lifetime images in order to check for potential artefacts.

Edge artefacts are always possible. At the boundary between of fluorescent
object and the background, you will have a decrease of signal-to-noise ratio
and signal-to-background. This may concur to skew/bias your fitting results.
Combine this with binning and you may get nice images but with edge
artefacts.

Not sure if this is relevant to your issue though...

Cheers,

Alessandro
www.quantitative-microscopy.org

[Aleem]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Alessandro,

I am trying to understand the effect of binning on lifetime resolution. I
understand that if we bin our lifetime data we combine information from
neighboring pixels, suppose if we have two pixels side by side with and
without signal and we bin the data, the final lifetime image will shows the
same information that bright pixel will have a finite lifetime whereas dark
pixel won't. This isn't clear to me, why the dark pixel having no effect of
neighbor bright pixel?

Thanks,

Aleem


On Fri, Apr 26, 2013 at 6:50 AM, Alessandro Esposito <
[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Dear Aleem,
>     perhaps it would help if you specify in more detail which is the issue
> you are
> trying to explain.
>
> However, a general note of caution when using the "binning" option in
> SPCImage is necessary.
>
> First of all it is not technically binning. It is a convolution averaging
> filter. It is
> slower than binning, but a good idea as it permits to improve SNR with a
> better
> compromise in respect of loss in spatial resolution resolution.
>
> Second, the bin value you input in the GUI is not the dimension of the
> kernel.
> The kernel dimension is 2*x+1.
>
> Therefore, who does not read the manual, with x=2 may think to bin 2x2
> pixel
> together, but actually this would be an averging filter with a kernel of
> 5x5
> pixel.
>
> Third, the default option of visualization in SPCImage is to overlay
> lifetime
> maps with intensity maps. Beautiful representation that provide the
> advantages of a smoother lifetime image (when using "binning") with the
> perception of higher resolution provided by the overlay with the non-binned
> intensity image.
>
> All legit and described by B&H. However, users should always inspect the
> "raw"
> lifetime images in order to check for potential artefacts.
>
> Edge artefacts are always possible. At the boundary between of fluorescent
> object and the background, you will have a decrease of signal-to-noise
> ratio
> and signal-to-background. This may concur to skew/bias your fitting
> results.
> Combine this with binning and you may get nice images but with edge
> artefacts.
>
> Not sure if this is relevant to your issue though...
>
> Cheers,
>
> Alessandro
> www.quantitative-microscopy.org
>

[Unruh, Jay]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Aleem and Alessandro,

I agree with everything that Alessandro has said so far.  Edge effects happen (at least for me) because background regions always have some signal and the background lifetime is almost always much shorter than my sample's lifetime (perhaps SHG, media autofluorescence, scatter, ect).  My general solution to edge effects has been to threshold or select regions first and then bin only within thresholded regions for analysis.  Of course that eliminates the possibility of a pretty lifetime overlay, but those are probably only for cover art anyway:)  Of course all of this eliminates the possibility of using SPCImage, but I quit using it a long time ago when I couldn't get it to display my raw curves on a linear scale (low photon count data always looks better than it should on a log scale)!  Perhaps they fixed that since then.

I have also found that for small Gaussian shaped objects with uniform background, a good option is to get the average center and standard deviation and fit the Gaussian in each channel image to get the amplitude and baseline.  If you have enough photons, this method naturally eliminates the background.  A good way to explore for background issues without fitting is the phasor method that Enrico Gratton has published on.  There you can quickly visually correlate pixels approaching the background with reduced lifetimes.

You will have to contact B&H for import options and file specifications.  I wrote an sdt importer that I can share with you, but it is a bit of a hack and I doubt it can be used for export.  If you contact B&H, perhaps they have some sample code for conversion.  I think there is an SDK for the acquisition software--perhaps that is a good place to start.  The folks at Boston Electronics have always been helpful for me so I would contact them as well.

I also agree with Alessandro that matlab is probably your best option for simulation.  For the complexity of what you want to do, you will need some sort of custom programming and matlab is the quickest way to start (though perhaps the slowest at running your code).  Don't forget to use poisson photon counting noise.  That can make a big difference for FLIM analysis.  I wouldn't worry too much about the monte carlo stuff unless you are doing single molecule or excited state dynamics.

Jay

-----Original Message-----
From: Confocal Microscopy List [mailto:[hidden email]] On Behalf Of Aleem Syed
Sent: Friday, April 26, 2013 8:25 AM
To: [hidden email]
Subject: Re: Simulation of a fluorescence lifetime image with fixed lifetime in one pixel

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Alessandro,

I am trying to understand the effect of binning on lifetime resolution. I understand that if we bin our lifetime data we combine information from neighboring pixels, suppose if we have two pixels side by side with and without signal and we bin the data, the final lifetime image will shows the same information that bright pixel will have a finite lifetime whereas dark pixel won't. This isn't clear to me, why the dark pixel having no effect of neighbor bright pixel?

Thanks,

Aleem


On Fri, Apr 26, 2013 at 6:50 AM, Alessandro Esposito < [hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Dear Aleem,
>     perhaps it would help if you specify in more detail which is the
> issue you are trying to explain.
>
> However, a general note of caution when using the "binning" option in
> SPCImage is necessary.
>
> First of all it is not technically binning. It is a convolution
> averaging filter. It is slower than binning, but a good idea as it
> permits to improve SNR with a better compromise in respect of loss in
> spatial resolution resolution.
>
> Second, the bin value you input in the GUI is not the dimension of the
> kernel.
> The kernel dimension is 2*x+1.
>
> Therefore, who does not read the manual, with x=2 may think to bin 2x2
> pixel together, but actually this would be an averging filter with a
> kernel of
> 5x5
> pixel.
>
> Third, the default option of visualization in SPCImage is to overlay
> lifetime maps with intensity maps. Beautiful representation that
> provide the advantages of a smoother lifetime image (when using
> "binning") with the perception of higher resolution provided by the
> overlay with the non-binned intensity image.
>
> All legit and described by B&H. However, users should always inspect
> the "raw"
> lifetime images in order to check for potential artefacts.
>
> Edge artefacts are always possible. At the boundary between of
> fluorescent object and the background, you will have a decrease of
> signal-to-noise ratio and signal-to-background. This may concur to
> skew/bias your fitting results.
> Combine this with binning and you may get nice images but with edge
> artefacts.
>
> Not sure if this is relevant to your issue though...
>
> Cheers,
>
> Alessandro
> www.quantitative-microscopy.org
>

[Alessandro Esposito]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear Aleem,
   in response to "why the dark pixel having no effect of neighbor bright pixel",
if a dark pixel is really dark, it does not contribute to the signal and therefore
it will not have any effect. Why should be otherwise?

Jay expanded on the possible sources of noise or background signal in
the "dark pixel". They are never really dark and therefore they will always have
an effect (small or large) on the analysis.

A note on the phasors. I guess I wrote enough in papers in support of phasors
or similar transforms, therefore do not misunderstand the following. However,
scatter plots of lifetime vs intensity may be more beneficial in this case.
Noticing correlations between intensity and lifetime, unless biologically
relevant, can indicate issues of the type we are discussing or pulse pile-up. In
the presence of of poor SNR/SBR, lifetime vs intensity may be a bit more
intelligible.

Cheers,

Alessandro

[Unruh, Jay]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

I agree about phasors.  The best thing about phasors is the ability to interactively "browse" your data, getting some sense of where similar lifetimes exist in the image.  For that matter, there is nothing magical about the fourier transform other than that its components are additive and have some established relationship to lifetime.  The method of moments can be made to behave very similarly with the same additivity.  The key is to find simple ways to visualize spatial relationships in your sample (preferably independent of complex fitting) so that you aren't just blindly trusting what the multiexponential fit gives you.

Jay

-----Original Message-----
From: Confocal Microscopy List [mailto:[hidden email]] On Behalf Of Alessandro Esposito
Sent: Friday, April 26, 2013 10:23 AM
To: [hidden email]
Subject: Re: Simulation of a fluorescence lifetime image with fixed lifetime in one pixel

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Dear Aleem,
   in response to "why the dark pixel having no effect of neighbor bright pixel", if a dark pixel is really dark, it does not contribute to the signal and therefore it will not have any effect. Why should be otherwise?

Jay expanded on the possible sources of noise or background signal in the "dark pixel". They are never really dark and therefore they will always have an effect (small or large) on the analysis.

A note on the phasors. I guess I wrote enough in papers in support of phasors or similar transforms, therefore do not misunderstand the following. However, scatter plots of lifetime vs intensity may be more beneficial in this case.
Noticing correlations between intensity and lifetime, unless biologically relevant, can indicate issues of the type we are discussing or pulse pile-up. In the presence of of poor SNR/SBR, lifetime vs intensity may be a bit more intelligible.

Cheers,

Alessandro