What is FastLifetime?
TL;DR: Fast lifetime is a fast way to estimate average lifetime per pixel without fitting,
but taking into account the IRF (instrument response function).
It is basically a "honest" average lifetime of photons per pixel, minus IRF(t=0) estimate.
IRF(t=0) is estimated differently in different software (by this plugin, SymPhoTime, Leica, Zeiss),
therefore fast lifetimes can be a little bit different, up to some constant additive factor (usually small).
The main goal of FLIM is to characterize lifetime decay curves.
This is best achieved by curve fitting and extracting the relevant parameters.
However, curve fitting is relatively complex and requires the choice of a model and parameters.
So at some point (I think) in the PicoQuant SymPhoTime software appeared so-called "fast lifetime"
as a way to estimate IRF "corrected" average lifetime.
What is IRF, and why do we need to correct for it?
IRF stands for Instrument Response Function.
I will explain it using a specific example.
So the lifetimes of fluorophores are distributed exponentially.
For now, let's consider a single, "monoexponential" distribution of a decay
with a single "average lifetime" decay constant.
If we were to measure photon arrival times in an "ideal world"
and then just calculate its average,
it would be the same as the exponential fitted lifetime (it is a property of the distribution).
In reality, this would not be the same,
because we measure it with an instrument, which is not ideal or perfect.
Our instrument (microscope + counting hardware) has so-called IRF or instrument response function.
Even if the photon arrives immediately, right after the excitation pulse,
we cannot detect its time as zero. The instrument will detect it with a delay and a bit of a "spread".
Because of many things (shown below), this "immediate" photon time distribution is gonna look something like this:
This is the IRF that (in principle) can be measured.
Taking this into account, our "ideal" exponential distribution
in the process of measurement is also going to change its shape.
It is said that the distribution becomes convoluted with IRF.
And so the "honest" average lifetime would not be equal to
the exponential decay coefficient.
In short, it is going to be larger (since we are omitting short lifetimes near zero).
So, to get a better, more precise "true" average lifetime estimate, it is preferable
to take into account the effect of IRF.
Usually, it is done by fitting: "tail-fitting" (ignoring the first part),
reconvolution or deconvolution, etc.
Again, these approaches are:
- Relatively computationally heavy
- Require many more photons, since we need enough observations to see the shape of the distribution
- We need to know IRF (measure it or estimate it somehow from the data itself).
Alternatively, there is a simpler approach for a better approximation
of the average lifetime and IRF. To my knowledge, its best description
is given in this chapter, paragraph 3, p.238.
M. Patting (2008) Evaluation of Time-Resolved Fluorescence Data: Typical Methods and Problems U. Resch-Genger (Ed.), Standardization and Quality Assurance in Fluorescence Measurements I: Techniques, Springer, Berlin Heidelberg, Berlin, Heidelberg (2008), pp. 233-258, 10.1007/4243_2008_020
In short, we can subtract IRF(t=0) from the "honest" average lifetime of photons,
which (especially in the case of monoexponential decay) will give us a good "corrected" measure.
This is usually what is shown as "FastLifetime" image in SymPhoTime and other software.
This approach does not require curve fitting,
it needs just a single parameter estimate of IRF(t=0) (average IRF lifetime).
That could be done without a separate acquisition, using raw data,
by detecting the steepest increase point in the rising part of the IRF convoluted decay curve.
This is according to the SymPhoTime documentation.
How exactly it is done in this software, I do not know,
but this is how I do it.
To estimate the "honest average lifetime" correction value IRF(t=0),
the plugin first builds a total histogram of all photons from all frames per channel.
Then we calculate its derivative (pointwise difference)
In principle, the peak position of the derivative should correspond
to the steepest increase. But we refine this value a bit more using neighboring values
and the formula for the "centroid of spectral lines".
The fitted value is reported in the Log window, see:
Estimated IRF t=0 for channel 0: 1.2967138ns
In principle, there are better (but more complicated) methods
to "blindly" estimate IRF (see this paper, for example).