1: iScope Microscope Design and Assembly
General comments on timing.
In Year 01 of the
grant, we will build two new but relatively inexpensive Internet
microscope systems with ample free space above the specimen. Drs. Williams
and Nissanov will work on integrating the carousel or jukebox
slide-handling system (Aim 2) with the microscopes. One system will be
based on an inverted microscope, the other either on another inverted
stand or on a fixed-stage upright microscope. In Year 03 we will acquire
two additional microscope stands with the aim of eventually putting five
fully configured iScopes on the web. This is a realistic schedulethe
first iScope was assembled over a two-week period. The major delay was in
having custom adapters made for the Canon XL-1 camera.
Most research microscopes are now, in
our opinion, overpriced and over-equipped. We will continue to do our best
to provide top-notch image quality on a relatively modest budget. We will
design and construct custom microscope parts and video adapters. When
practical, and when the price and quality are right, we will purchase used
microscope stands, in particular older-generation Zeiss ICM and Universal
stands. We have already purchased most of the optics used by Zeiss
Axioplan microscopes, and we will therefore buy at least one of these
microscope stands. They have an extraordinarily precise focusing mechanism
based on a "harmonic" drive that has almost no mechanical
backlash. Our major expenses will be motorized stages, optics (condenser,
an objective, and Normarski prisms), digital video cameras, and fast
microcomputers. These "peripherals" are actually the key pieces
of equipment for this project. Collectively these parts cost approximately
$25,000 for our first iScope. We can purchase stable microscope stands
with good focusing blocks for about $10,000. Thus, over the course of this
grant we expect to spend approximately $140,000 on the microscopesless
than the cost of a single confocal microscope system. All four
microscopes, along with our exiting iScope, will remain online after the
end of funding.
iScope Feature Set.
The following is a short
list of key features of the iScopes.
For streaming video by web microscopists, we are leaning toward use of a
single x40 dry objective (NA 0.65) with proven optical resolution of 0.8
µm and pixel resolution as high as .1 µm/pixel. We will experiment with
higher-NA dry and water immersion objectives for microscopes intended
specifically to acquire very high resolution Z-axis image stacks. The
higher the NA, however, the more difficult the implementation and the
greater the maintenance problems.
2. For streaming video, two magnifications
will be provided simultaneously via dual 3-CCD DV cameras, one with a
field of view of 60 x 80 µm, the second with a field of 200 x 00 µm.
3. DC servo motorized stage and motorized Z
axis adapted for the carousel slide delivery system.
4. Differential interference contrast optics.
6.Alternative light sources: a low-voltage
tungsten bulb with a very extended life for streaming web microscopy and a
high-pressure Xenon-Mercury system (1000 h bulb life) for acquiring Z-axis
Justification for DIC optics of Nissl-stained
tissue. Standard brightfield microscopy of Nissl-stained specimens
primarily conveys information about the optical density of the strain. DIC
optics reveals differences in the refractive properties of the tissue. DIC
optics of stained tissue is thus a multimodal imaging method and as such
provides more information about the tissue (Farkas et al., 1995; Glasbey
and Martin, 1996) than simple brightfield microscopy. A number of other
advantages of DIC microscopy have been enumerated in Williams and Rakic,
1988a,b). One of the most important has to do with the unique modulation
transfer function of DIC images (Inou, 1986). In short, DIC optics
enhances contrast of high-spatial-frequency image content and suppresses
contrast of low-spatial-frequency components of the imagebasically, DIC
acts as a high-pass optical filter and edge enhancer. Furthermore, the
depth of field of the high spatial frequencies is remarkably shallow,
making it possible to optically section DIC through-focus series far more
effectively than corresponding non-DIC images. DIC can be thought of as a
pseudo-confocal system in which the out-of-focus blur from tissue above
and below the focal plane is significantly attenuated (Oldenbourg et al.,
1993). It is still not perfect, and many of the problems in
high-resolution reconstruction of single cells discussed by Hibbard et al.
(1996) remain. ABRUPT CHANGE HERE> The contrast enhancement of DIC also
makes it possible to fully open the condenser aperture to match the
maximum aperture of the objective. The usual compromise between high
numerical aperture and low contrast is thereby avoided. There are no
significant countervailing disadvantages of DIC except possibly the cost
involved in adding it to the microscope.
Autofocusing the microscope.
efficient autofocusing methods for the iScope is critical both when the
microscopes are being controlled during a streaming video session and
during the capture of through-focus series (Aims 3 and 4 below). Focus
would not be quite so critical if the slides were not so large and if the
sections were perfectly flatly pressed to the slide surface.
Unfortunately, and the difference in the z-axis coordinate of the top of
different sections on a single slide can be up to 100 µm. Thus, without
automatic z-axis focusing it will be very difficult for our clients to
remain in focus when leaping from one section to the next. What are some
possible solutions? Connecting the XL-1 cameras internal autofocusing
system with our motorized fine-focus may be an inexpensive and elegant
solution. This is particularly practical with the Canon XL-1 because all
lens controls, including those that drive the automatic lens focus, are
readily accessible (the autofocus lens dismounts on this camcorder). Focus
in the XL-1 is driven by the image quality itself, not by an infrared
system. John Zemek, president of Applied Scientific Imaging, has offered
to help us test the feasibility of this approach. A standard fallback is
to use image analysis utility programs to access image quality and
independently dive the z-axis. For example, Wu et al. (1996) at Mt. Sinai
School of Medicine have developed efficient methods to drive an autofocus
mechanism. More than 10 autofocusing algorithms that we will consider are
succinctly reviewed by Santos et al. (1997).
DIC and focusing.
One of the minor advantages
of DIC optics is that even in completely unstained parts of the section
there will be ample phase contrast to generate the high-spatial-frequency
signals of a well-focused image. When the image is not over tissue, but is
over a blank part of the slide, we do not want the focusing mechanism to
hunt forever for focus or to focus on the top of the cover glass. A
possible solution to this problem is to force the focus control through
one cycle of the maximum z-axis travel permitted for the stagesay
200 µmand if no z coordinate with adequate focus is detected, rest the
microscope at the last focused z-axis coordinate. If the lag in control of
the fine-focus can be reduced to a sufficiently short interval of < 1
second between command and response, then it may be easier simply to
disengage the autofocus mechanisms entirely.
The PI has spent a
considerable amount of time working on this problem. There are several
vertical calibration standards in the form of series of polymer films of
very precisely known thickness. However, these films have refractive
indices that differ from glass and oil (~ n = 1.52), adding complications
to the vertical calibration process. Boddeke et al. (1997) have developed
a simple and elegant method to calibrate the z-axis using a "tilted
slide" preparation. The slide can be a standard calibration slide
with a series of closely spaced parallel lines. Moving the tilted slide
changes the position of the line that is in best focus as a function of
the tangent of the tilt angle. Their protocols permit estimates of
backlash in the focusing block, long-term stability, and absolute
Several factors affect
how accurately a given coordinate on any slide can be acquired and then
reacquired. One factor is the adequacy of the database that describes the
slide and the section coordinates. Assuming that the database is correct,
then there are a series of physical factors that affect relocation
accuracy: how precisely the slide is positioned on the stage, the
consistency of encoding on different microscopes, etc. Tucker et al.
(1994) achieved a remarkable relocation precision of less than 17 µm
between two different microscopes and a precision of less than 7 µm on
the same microscope. R. Williams has constructed an encoded stage using a
trio of Heidenhain digital length gauges that can consistently reposition
a manually placed slide to within 5 µm radius. The ASI DC servo-motor
stage that is now fit to the iScope has a relocation accuracy of better
than40 µm even when the microscopist wanders all over the 2 x 3 inch
slide prior to centering.
For the purposes of this research project, we will initially be satisfied
if a client can select a point on one of our low-resolution slide images
(25 µm/pixel) and then be delivered to that coordinate within 100 µm on
any of the microscopes. This will require a database with an entry for
each slide. This Slide-Coordinate database (described in Project 1) will
be used to navigate from images in the MBL to points on particular slides.
Slide registration on the microscope stage must be precise, and this
registration will be improved by the use of the holders that will be
designed with this problem in mind. We ultimately hope to attain
relocation accuracy of better than 00 µm. The main purpose is to deliver
users to single regions that were previously used to generate
through-focus series. Relocation accuracy of 20 µm would be required
to put the user in a field that overlaps the initial images. Having this
level of precision will be useful to confirm and extend datasets generated
using only MBL image resources. With the use of absolute slide coordinates
and the ability to rezero each slide after it is loaded, we should be able
to achieve good relocation precision.
A single objective will
be used in each iScope, simplifying relocation accuracy because we will
not have to worry about nosepiece position and centering multiple
objectives. However, we will be using multiple iScopes, and it will be to
measure offsets of each objective on each microscope so that systematic
bias in location among microscopes is minimized. Using a single objective
per microscope obviously simplifies this procedure. Each microscope may
require its own offset.
Software used in conjuction with the iScopes.
The current user interface for the iScope is inadequate in many ways. The
principal problem is that it is very difficult to navigate across a very
large slide (50 x 75 mm) at a magnification at which the field of view is
less than 80 by 100 µm. We have three ideas that should greatly improve
navigation and which for the most part will be easy to implement:
1. Simultaneous display of a
low-magnification image of the entire slide (taken from the MBL) with
easy-to-read coordinates, or better yet, a cross-hair that corresponds to
the site of the high magnification site being imaged by the iScope.
2. Greatly improved response time between
initiating a movement and a video refresh that reflects that change.
iScope use arbitration.
implementation is a simple first come, first served system. Any user can
work or play with the microscope for a 10-minute period. Registered
scientists actively using the MBL and iScope will be able to sign in for
periods of 2 hours or more, depending on demand. We will come back to the
issue of access in Section 4, but in short, we do not expect access
problems because two microscopes will be on reserve specifically to
acquire high-power image stacks at locations requested by our clients. It
should be possible to define these locations using the 4.5-µm pixel
images. Thus, even heavy users should not need long periods of
uninterrupted access to the microscopes.