>Thanks to Guy for his very helpful and complete description of how he
>measures "resolution" (a term I find very slippery, though nonetheless
>necessary).
Absolutely! Rayleigh resolution and FWHM are not the same thing, and
various other parameters, such as the precision with which we can locate
objects (recently discussed in this group) are different things entirely.
The main merit of FWHM, my favoured parameter, is that it is relatively
simple to measure. Also - see below - in practice one can resolve apart
separate objects separated by the FWHM.
>
>I think that it is important to remember that the "size" of the data
>recorded in this fashion is not strictly speaking the instrumental "Point
>Spread Function" (PSF) that appears on all of the theoretical charts and
>graphs, but the PSF convolved (read as: multiplied-in-3-dimensions) with
>the ACTUAL size of the bead. Larger beads will give "worse" resolution.
>Assuming that one is using 100nm beads, you are ALMOST justified in
>subtracting 100 nm from your measured FWHM.
I'm glad you said ALMOST, Jim! As Stefan Hell has already pointed out,
for practical purposes 100nm is small enough to be regarded as a point.
You are certainly NOT justified in subtracting 100nm from the measured
FWHM! It is, of course, true that in principle the size of the actual
bead wil have _some_ effect on the measured image, and last night I tried
out some quick tests to find out how much.
I ran two samples - Biorad's standard 210nm bead slide supplied with their
systems and 100nm beads (thanks to Carol Cogswell for these) in Cityfluor.
The lens used was a x63 Plan Apochromat NA 1.4.
Lateral FWHM with the 210nm beads was ~ 300nm, Axial ~ 700nm. The vertical
profile looked a bit asymmetric, implying that the mounting medium (can
anyone from Biorad tell me what it is?) was not a perfect match. There
were quite a few beads fairly close together, and beads 280 nm apart
(centre to centre) were easily resolved.
With the 100nm beads lateral FWHM was ~ 240nm and axial was 400 - 450nm
(it's hard to be more precise when you're working in 180nm steps!). The
profile looked reasonably symmetrical though the side lobes were quite
noticeable (I had the background fairly high).
So adding 110nm to the bead diameter has only increased the profile
(laterally) by ~60 nm. And even that may be largely due to refractive
index mismatch. This implies to me that reducing it below 100nm would
have a pretty insignificant effect.
>Two remaining questions:
>
>1. Guy, could you tell us about the rest of your system? Does it include
>a "normal" fluorescence attachment or a DIC module? As I remember, on the
>Optiphot, both of these impose a 1.25x magnification increase when viewed
>from above and causes an inverse effect on the size of the beam in the BFP
>which may explain some of the differences in resolution discussed here.
Confocal microscopes didn't exist when the Optiphot was a current model!
My only recent experience with 160mm tube-length Zeiss systems is on a
12-year-old Zeiss Standard, and as Jim implies, putting anything into the
tube is an optical disaster. The fluorescence attachment is not properly
compensated and when we once tried putting a confocal head on it (the
little HBH, predecessor of the Optiscan) we had to take out the fluorescence
attachment to make it work. So I wouldn't advise anyone to add confocal
to an ancient microscope.
We use a Zeiss Axiophot, which is an infinite tube-length system so that
whatever you put into the tube makes no difference. However you don't
want too much glass in the way, so I don't advise adding anything that
can't be pulled out to give a clear path. And one thing which is absolutely
essential is that you pull out any DIC prisms before doing confocal work.
The laser beam is inherently polarised and they will completely screw up
your resolution. (One friend of mine - no names, no pack drill - actually
called in the company engineer because his new confocal was giving lousy
resolution - and had to be gently reminded to pull out the DIC prism ...)
>
>2. Do you have any problem with fading?
No, I don't. Both the Biorad-supplied 210nm bead slide and my Molecular
Probes 100nm beads in Cityfluor seem durable enough to record a z-series
at adequate sampling for a reasonable measurement. Sure, there's noise,
but you measure a few beads. The real limitations are the step size we
can get (too big) and the precision of the Zeiss focus drive (woeful).
I hope soon to be able to try a precision x-y-z stage which should give
much better sampling. Of course it may be that fading _will_ then become
a problem!
>As I assume the answer is yes,
>what do you do about it? (Laser power levels? Zoom/pixel-size/plane
>spacing? Techniques like "focus"-here/collect-data-there?
>Do you "fit" the
>data recorded to some sort of mathematical function in order to reduce the
>effects of noise on your measurement of FWHM, or just estimate my eye?
I don't, but given the limited sampling imposed by the Z-drive it would
in principle seem like a good idea. However you'd probably have to smooth
the data first and by then you're getting a long way from the original ...
So I just eyeball it and do quite a few beads so as to get an overall
picture.
>Other practical suggestions?
Yes, one. Doing a line scan doesn't, in practice, always quite hit the
line you've selected in the X-Y image. (The ballistics of the galvo
mirror, I presume). So you may need to take a few shots. The other
approach is to take a complete series of sections and extract an XZ
or YZ section from these. Then it's easy to get the profile you want
_but_ with most software (certainly with mine) you are limited to
integer interpolation so you'll have to go for the nearest figure and
make a small correction to your vertical measurements. My software
doesn't do linear interpolation, either - it just duplicates slices.
I regard this as preferable but it's open to argument. Voxel-View
does linear interpolation when it 'pads out' a volume, and you should
be aware of this if you use it to generate your XZ sections.
Guy Cox
|
