Hello Mario et al

I know very little about about optics other than what is discussed on
this list and a bit of reading around the well known texts... what
follows is based empirically on our experience, and I'll gladly take
advice on the way we are interpretting our results...

We use both confocal and electron microscopy to look at the dendritic
structure of neurons and their synaptic boutons. The boutons range in
diameter from about 0.1 to about 5 um in diameter and the neuronal
dendritic processes commonly have a similar diameter range. As I've
mentioned on the list a few times, we have watched what happens to our
cells as well dye fill them and take them through fixation, clearing,
mounting etc for both confocal and TEM imaging, so we know about the
dimensional changes that take place etc (not much as it turns out).

Our experience (and that of many others, judging from the published
pictures in the literature) is that a straight reconstruction of a stack
of confocal images taken with a range of nominal z-distances between
them results in processes being distorted in the z-axis so that they
look like ribbons, and the boutons are distorted into vertical
ellipsoids. This is clearly "wrong" as verified by EM (with due
allowance for fixation artefacts). If we use a blind decovolution
approach (AutoDeBlur), and then do our reconstructions, this z-axis
distortion is usually completely removed and the dimensions of our
structures match those seesm with EM. When we are trying to measure
surface areas of processes and distances from boutons to dendrites etc,
this differences really matter a lot.

There are other things that matter as well, such as differences in
thresholding level, surface noise etc, but all of these are easier to
manage and standardise after deconvolution...

So the question to Mario and others is this:
        Are we getting better "resolution"? Or are we just getting a better
defined image at the resolution we started with? And in the end, does
the difference matter?

Looking forward to nay replies!!

IAN



Mario Moronne wrote:
>
> Hi All,
>
> Forgive me if this question demonstrates laziness on my part (just
> had shoulder surgery and don't feel like flipping journal pages: poor
> me). Answers can be sent to me directly, or not.
>
> I am stuck with the notion that the resolution in any direction will
> be limited by the diffraction properties of the objective (and
> specimen-e.g., RI mismatch). For the ubiquitous 1.4 NA oil that I
> think goes unspoken as the point of reference, the FWHM distance in z
> with no spherical aberration using 488 nm light is something like
> 0.45 um. In practice, even staying near to the coverslip (~10 um) it
> is more like 0.8 um. However, let's forget the latter and stick with
> 0.45 um. Nonetheless, if we use the idea of overlapping
> distinguishable objects in z how can one achieve 0.2 um resolution
> through deconvolution?
>
> I have no doubt that deconvolution can help confocal z-resolution,
> but how does it get below the nominal diffraction limit? Is this
> meant to refer to visibility or formulaic calculation. I don't mean
> to dredge up arguments about what one can see as opposed to
> mathematical conventions such as the Rayleigh limit, but perhaps this
> is unavoidable. Okay, so in a simplistic fashion one must consider at
> minimum two things: the z-resolution stated in terms that are
> reasonably approximated by a SINC function which gives us the FWHM
> 0.45 um and the Rose Criterion which pertains to human recognition of
> differing contrast, telling us that for those not especially gifted,
> contrast difference must be 3 to 5 times larger than the noise of the
> objects being visualized. The 3-5 range depends on the geometry of
> the object and for overlapping objects in z, I am not clear what is
> the better number. Nonetheless, the issue of noise seems to me to be
> very important when contemplating the ability of a deconvolved
> confocal stack to obtain 0.2 um resolution or distinguishability. For
> one thing it implies that the noise level is reduced to very low
> levels. Even in conventional W.F. deconvolution, noise which is often
> many many time lower than in confocal, is an important factor in
> deriving good reconstructions; hence all the smoothing operations in
> most methods. Confocal is aided by the fact that deconvolution is
> already partly or nearly entirely completed. Still, how many people
> are using photon counting, which actually is representative of the
> statistical noise in the sample? How much averaging is required to
> get good enough contrast to give 0.2 um?
>
> As a practical matter, the ability of a program to provide 0.2 um z
> resolution of confocal stacks along with its application in typical
> W.F. decon. for me would definitely tilt me in the direction of a
> package like AutoQuant's. So, can it really do it and under what
> circumstances?
>
> Tim, if you have some example images on your website etc., I would
> really like to see them. Can you point me the way to profile plots or
> send me some jpegs demonstrating such z resolution? Overlapping
> microtubules, maybe? Thanks and hope that I am sounding merely
> ignorant rather than Grinchy.
>
> Happy Holidays,
> Mario
>
> >AutoQuant sells a deconvolution package for the confocal microscope and for
> >widefield optics.  www.aqi.com
> >
> >There are some papers on the topic you brought up about which modality gives
> >sharper
> >images.
> >
> >See the paper by Peter Shaw,   Prog. Biophys. molec. Biol., Vol. 56, pp.
> >187- 213, 1991.
> >and the chapter by me in The Handbook of Biological Confocal Microscopy, by
> >J. Pawley.
> >
> >Generally, our experience and as pointed out by Peter Shaw is that:
> >
> >Widefield deconvolution and confocal microscopy without deconvolution
> >provide
> >    comparable resolution in all 3 directions (x, y and z).  The widefield
> >case provides a little resoution improvement, but it is subtle. If you are
> >making it
> >    choice between these 2 modalities, lower noise (better signal to noise),
> >lower
> >         photobleaching, faster acquisition and some other advantages speak in
> >         favor of widefield.  The fact that no time is needed for deconvolution
> >         and that there are no deconvolution-induced artifacts speaks
> >in favor of
> >         confocal.
> >
> >Confocal with deconvolution provides far superior resolution, especially
> >along z,
> >compared to widefield with deconvolution.  Depending upon the settup, you
> >can expect on the order of 0.2 micrometers resolution along z, compared to
> >about 0.5 to 0.8 with widefield.
> --
> _____________________________________________________________________
> Mario M. Moronne, Ph.D.
> NanoMed Technologies
> FAX (510) 528-8076
> Berkeley, CA
> 94706
>
> [log in to unmask]
> [log in to unmask]

--
Professor Ian Gibbins
Anatomy & Histology
Flinders University of South Australia
GPO Box 2100, Adelaide, SA 5001
Australia

Phone:  +61-8-8204 5271
FAX:    +61-8-8277 0085
Email:  [log in to unmask]