Search the CONFOCAL archive at
http://listserv.acsu.buffalo.edu/cgi-bin/wa?S1=confocal

Wes,

I agree with your doubts. Being able to calibrate intensities is a good
thing, but variabilities in staining and background can be considerable
(even within one preparation!), and an absolute calibration of intensity
does not help here.
I also think that using the same threshold for two test samples might
actually bias the result towards one of them. In our group we work a lot
with interphase chromosomes, so let me use them as an example: Imagine
two homologous chromosomes in one nucleus stained by a FISH probe. They
should have the same DNA content, but one might be compact and the other
one more decondensed and extended. The first one is brighter and smaller
and the second one is dimmer and larger, but the total fluorescence
should be the same. When measuring their volumes (or diameters or
whatever) you have two sources for error:
1. The limited resolution of the light microscope blur the "borders" of
the chromosomes, and this will mostly make the brighter chromosome
larger. This can possibly be alleviated by deconvolution.
2. Faint signals can become indistinguishable from the background. This
will work against the dimmer chromosome making it smaller.
Now you can try to find a common threshold that makes the total
intensity for both chromosomes equal, but it might be that there is no
such threshold, because both errors give a bias towards higher volumes
for the brighter chromosome and lower volumes for the dimmer chromosome.
  So you might have no choice but to use different thresholds "by eye"
for the two homologous chromosomes (uaaargh - bias introduced by the
investigator). Usually a range of thresholds will lead to plausible
results, so my approach would be to quote the volumes measured with the
highest/lowest threshold as error ranges. That's again a source of error
(what's the "real" highest/lowest threshold?), but it's best you can do.
What do people think of this procedure? I know there are more contrived
algorithms around than simple thresholding. Can anybody comment on those?

Joachim



Wes Wallace wrote:

> Search the CONFOCAL archive at
> http://listserv.acsu.buffalo.edu/cgi-bin/wa?S1=confocal
>
> Dear confocalists,
>
> What I've gathered from the postings so far is that one should use
> reference standards at each imaging session to calibrate the brightness
> range of the imaging system for that session.  The reference standards for
> this purpose can be uniformly fluorescent slides, as suggested by Barbara
> Foster.
>
> Once this is done one must keep the same brightness and contrast settings
> on all images acquired in that session, so that all biological structures
> in the image will be imaged using the same conditions.  Furthermore, to
> avoid spreading and flattening of structures, one must avoid any saturated
> pixels in the images: therefore when setting brightness and contrast for a
> given imaging session, one should use both the reference standards and a
> particularly bright biological sample as well, to make sure there is no
> saturation.
>
> Finally one should acquire an image of a structure of known size, such as
> a fluorescent sphere.  One can then measure this structure in the acquired
> images to determine how large its apparent size is and calculate a
> correction to the measurements.
>
>
> COMMENTS:
>
> 1.  The use of uniformly fluorescent slides to calibrate the response of
> the imaging system seems like a very good idea, but there is still the
> question of linearity of response.
>
> Let me give an example: I have a reference slide which I assume to be of
> consistent brightness from session to session.  On my first session within
> a given experiment, I start by putting in the reference slide.  I adjust
> the dynamic range (using laser power, PMT sensitivity, offset, etc.) so
> that the slide gives an image with pixel values of 100 (out of 256) all
> across the image.  I now put in a biological structure.  The biological
> structure is much dimmer than the reference slide, so that the brightest
> features in the structure are giving pixel values of say 50.  I now
> increase the brightness so that the brightest features are giving pixel
> values of 250 (not saturating, but fully using the dynamic range).
>
> How do I compare this adjustment on different sessions?  It seems to me I
> should follow exactly the same procedure at each session, so that the
> brightness range would have to be adjusted once and for all on the first
> session.  But this would still assume that if I increase the brightness by
> 200 units after viewing the reference slide, this means the same thing on
> different days.  Is that a safe assumption?
>
> Also, what if the brightness of my biological structures varies from
> sample to sample due to staining variability or differences in background
> fluorescence?  I am only asking this question from a theoretical
> standpoint, I have not yet checked whether this kind of variation is
> really that prominent or whether it can be ignored as one post suggested.
>
> 2.  The use of fluorescent spheres to calibrate distance seems to me to be
> very problematic for the following reasons.
>
> First of all, optical distortions (such as diffraction, spherical
> aberration, etc.) which affect the measured size of a 6-micron sphere will
> very differently affect the size of a 1-micron object or a 15-micron
> object, let alone a non-spherical object.
>
> Secondly, even if I ignore this caveat, there is still the problem I
> mentioned in my first posting, of how to set the brightness and contrast
> settings in the acquired image.  When doing measurements on a computer
> screen one has a much smaller range of grey levels than is actually
> present in the data.  Therefore one must adjust contrast and brightness to
> optimize the visibility of a particular structure.  And in some cases one
> will be thresholding the structure into a binary image.   If the
> microsphere being used as a standard happens to be of a different
> brightness than the structures I intend to measure, I will be introducing
> additional distortions to the size of the measured structure.
>
> However, if I want to compare the size of some structure in two groups
> (say treated and untreated with some drug), I may not need to know the
> "true size".  In that case all that concerns me is that the structures in
> each group should have the same relative brightness, so that when I
> threshold them I am not introducing a bias towards one of the groups.  If
> I have calibrated the contrast and brightness settings, then the only
> remaining source of variability in brightness is the staining itself and
> background fluorescence. These may not be irrelevant factors!  But,
> assuming these factors are not biased towards one or the other
> experimental groups, it will be safe to ignore it if my only purpose is to
> measure a relative change.
>
>
> Thanks everybody, I hope this reasoning is on the right track
>
> Wes
>
>
> Wes Wallace
> Department of Neuroscience
> Brown University
> Providence, RI 02912 (USA)
> [log in to unmask]
>


--
------------------------------------------------------------------
   Joachim Walter, Dipl. Phys.
   Institut für Anthropologie und Humangenetik der LMU München
   AG Cremer
   Richard-Wagner-Straße 10/I
   D-80333 München               Tel. +49 - 89-2180-6713
   Germany                       Fax  +49 - 89-2180-6719

   [log in to unmask]
------------------------------------------------------------------