NAME
conime - analyze gel electrophoresis (RLGS) images of DNA
SYNOPSIS
conime [ file1, file2, ... ]
DESCRIPTION
Conime is a program to view, process and analyze two dimen-
sional gel electrophoresis images of DNA sequences, also
known as Restriction Landmark Genomic Scanning (RLGS) pro-
files. Files file1, file2,... contain the images in tiff
format. Supported operations include filtering and segmen-
tation, image registration, spot identification, spot meas-
urement and matching. Markers can be added to the images,
both for image annotation and for image registration. Spot
editing functions are also available.
DISPLAY AND DISPLAY OPERATIONS
Conime displays images in both single image viewports and
composite viewports. Single image viewports contain a sin-
gle gray image. Composite viewports contain colored over-
lays of multiple images or colored differences of two
images. An arrangement of viewports is called a view.
Display operations include scaling, translation, shading and
gray intensity normalization. Display operations effect
only the visualization of the image and do not have any
effect on image processing, registration, identification or
matching. The only exception is the location of images in
the composite viewport is used as an initial position for
image registration.
To translate a viewport across an image use either the mouse
buttons or the keyboard arrows. Pressing the left mouse
button selects the current viewport (highlighted in green).
Pressing the center mouse button centers the viewport on the
cursor location. Pressing and releasing the right mouse
button translates the image the distance and direction moved
between button press and button release. The center and
right mouse buttons only operate when the cursor is in the
current viewport. The keyboard arrows translate the current
viewport in the given direction by the translation step
size. The translation step size can be modified in the
translation dialog. If the current viewport is a composite
viewport, then all images in the composite viewport are
translated by exactly the same distance and direction.
Translating an image in one viewport causes the image to be
translated in all viewports containing the image. When the
lock toggle is on, translating an image causes all images in
all viewports to be translated by exactly the same distance
and direction. Registering images causes the lock toggle to
be set on.
A simple shading dialog allows users to increase the inten-
sity and contrast of the image for visualization purposes
only. Increasing lighten increases the mapping of light
pixels to white while increasing darken increases the map-
ping of dark pixels to black. Increasing both lighten and
darken increases the image contrast. When lighten and
darken are set to 0, the original pixel intensities are
used. Note that shading has no effect whatsoever on any of
the image processing tools.
Shading of filtered images can also be done using normaliza-
tion. (See SPOT MEASUREMENT AND MATCHING.) Display normal-
ization is for visualization purposes only and has no effect
on spot measurement and matching. Normalization parameters
are specified in the normalization parameters dialog under
the tools menu.
IMAGE INFORMATION
In image viewing mode, pressing the left button in the
current viewport prints out information about the pixel at
the current location. Pressing <Shift> and the left button
in the current viewport prints out information about the
spot at the current location.
Image information such as image size, number of spots,
number of identified spots, etc., is also available in an
image information dialog.
BACKGROUND FILTERING
Background filtering separates foreground from background
pixels. Parameters are the maximum spot cluster size, the
maximum noise size, and an opening size. Light pixels are
repeatedly removed from foreground until the number of pix-
els in every connected component of foreground pixels is
less than the maximum spot cluster size. The foreground is
opened by applying erosion with the given opening size fol-
lowed by dilation with the given opening size. Finally,
small components with size less than the maximum noise size
are set to background.
Increasing maximum spot cluster size causes more pixels to
be identified as foreground. Increasing opening size causes
more pixels to be removed from foreground in the opening
procedure. Increasing maximum noise size causes more pixels
to be identified as noise and removed from foreground.
CONTOUR SEGMENTATION
Foreground pixels are segmented into spots by identifying
small dark regions forming spot centers and adding a spot at
each center. Parameters are minimum spot center size and
minimum spot center gray depth. Light pixels are repeatedly
removed from foreground until the number of pixels in each
connected component is approximately the min spot center
size or the gray values of the pixels in a connected com-
ponent are within minimum spot center gray depth of each
other. Each connected component becomes a central region
for a spot and the spot center is located at the center of
the component. Increasing minimum spot center size or
minimum spot center gray depth causes foreground to be seg-
mented into fewer spots.
The parameter margin width is the width of a margin (in pix-
els) around the boundary of the image. No spots are created
whose centers lie in this margin.
TEMPLATE SEGMENTATION
After foreground pixels are segmented into spots, additional
segmentation can be performed by matching the image with a
template image and using the template image to guide further
addition or splitting of spots. Spots can also be identi-
fied and labelled by matching them with identified and
labelled spots from the template image.
After the initial matching and template segmentation, the
segmentation image can be reregistered with the template
image based on the spot matches and further segmentation can
be attempted. This matching, segmentation and reregistra-
tion can be iterated a number of times.
The segmentation and template images must be registered
before applying template segmentation. Thus an unfiltered
image must first be filtered and initially segmented, then
registered with the template and only then can it be seg-
mented with the template.
The segment dialog permits the choice of contour segmenta-
tion or template segmentation or both. Contour segmentation
reapplies segmentation based on contour area and has the
parameters as described above. Template segmentation
requires a filtered, segmented template image from which to
apply segmentation. Template segmentation parameters are
number of iterations, identify spots flag, and tag spots
flag. Number of iterations is the number of iterations of
the matching, segmentation and reregistration procedure. If
the identify spots flag is true, then spots on the segmenta-
tion image are identified and labelled using identified and
labelled spots on the template image. If tag spots flag is
true, then spots on the segmentation image have tags added
which indicate whether the spot was added or split and how
the spot was matched with a spot in the template image.
IMAGE REGISTRATION
Image registration translates and warps images so that they
best "match" the first image. Registration markers are
added to each image, and images are translated and warped to
align these registration markers. Registration markers can
either be computed by conime or added manually. Correspond-
ing identified spots can also be used to registre images.
Corresponding registration markers or spots which would
create large angular distortion in the images are ignored.
Conime computes registration markers by choosing a set of
locations in one image and attempting to find for each
chosen location a corresponding matching location in another
image. Two locations "match" if the distance between fore-
ground pixels in windows around the two locations is small.
The partial Hausdorff distance is used to measure the dis-
tance between the foreground pixels in the two images.
Matching may only be applied after the images are filtered.
Matching is a time consuming operation and should usually be
done on compressed versions of the images.
Registration parameters are the maximum match distance, the
match window width and height, the maximum number of regis-
tration markers, the percentage of pixels to ignore, the
matching threshold, the maximum angular distortion, the
registration marker group and the compression. Conime first
chooses a set of locations in each image. The maximum
number of registration markers is the size of those sets.
Conime then attempts to match locations in one image with
locations within the maximum match distance in the other
image. For images in the composite viewport, the distance
between the two locations in the composite viewport is used.
Therefore, the relative position of images in the composite
viewport can affect registration. Images which are not in
the composite viewport are assumed to be aligned on their
lower left corner. Conime uses the result of the initial
matching to suggest a corresponding location for each loca-
tion from the first image and search the neighborhood of
each corresponding location for a good match.
Increasing the maximum number of registration markers will
increase the number of registration markers created but slow
down the registration algorithm. Increasing the maximum
match distance may help conime register greatly misaligned
images, but will greatly slow down the algorithm.
The match window width and height determine the window width
and height used for comparing two locations. Increasing
this value will slow the algorithm and will change the set
of registration markers created.
The percent of pixels to ignore is the percent of pixels
ignored in computing the partial Hausdorff distance. Ignor-
ing these pixels allows conime to register images with added
or missing spots or significant image distortion.
Increasing this value will usually, although not neces-
sarily, increase the number of registration markers created.
The matching threshold determines the quality of a match.
The algorithm computes the partial Hausdorff distance
between two windows and then the expected partial Hausdorff
for a random distribution of foreground pixels in the win-
dows. Registration markers are only placed when partial
Hausdorff distance divided by the expected random partial
Hausdorff distance is less than the threshold. Increasing
this threshold increases the number of registration markers
created.
The registration marker group is the group label for the set
of registration markers created by the algorithm. Register
images deletes all current markers in the group before
creating the new set of registration markers. To register
an image with two other images, you need to register and
create registration markers for all three images simultane-
ously or use two different groups of registration markers,
one for each registration.
Maximum angular distortion is the maximum distortion permit-
ted between registration markers. The directional vector
between two locations in one image is compared with the
directional vector between the two corresponding locations
in the other image. If the angle between these vectors is
more than the maximum angular distortion, then registration
markers are not created for one of these corresponding loca-
tions. Note the underlying assumption that the vertical and
horizontal directions of all RLGS images is essentially con-
sistent. Increasing the maximum angular distortion will
increase the number of registration markers created by the
algorithm.
Compression is the compression used to register images.
Using no or less compression significantly slows the algo-
rithm.
SPOT IDENTIFICATION
Identified spots in one RLGS gel can be used to identify
spots in another gel. The two gels must be registered,
spots are matched, and then labels of spots in one gel are
copied to the corresponding spots in the other gel. The
identification procedure is conservative. Spots only match
if their centers are close together, if their pixels signi-
ficantly overlap and if the surrounding foreground pixels
match.
SPOT MEASUREMENT AND MATCHING
Spots can be measured individually or all spots can be meas-
ured simultaneously. Images should be registered before
measuring spots. Spots are measured by first normalizing
the gray scale of a window around the spot. Conime computes
the median background intensity and median spot intensity in
a neighborhood of the normalized spot and remaps gray scale
intensities, sending the background intensity to 0 and the
median spot intensity to about 6.0. It similarly normalizes
a corresponding window in every other image. Conime com-
putes the top 10% and median intensities of the spots pixels
and the top 10% and median intensities of the corresponding
pixels in every other image. Note that measurements are
pixel based and rely upon correct registration, not correct
spot segmentation, identification or matching.
Conime also provides the capability to search and mark all
added or missing spots in an image. A markup image is com-
pared with a reference image. An "added" marker is placed
over any spots in the markup image which are significantly
more intense than the corresponding pixels in the reference
image. A "missing" marker is placed over any locations in
the markup image which correspond to spots in the reference
image which are significantly more intense than spots in the
markup image. The difference and the ratio of spot intensi-
ties are used to compare spots.
CURRENT LIMITATIONS
-- Conime is extremely memory intensive using approximately
250Meg per 20Meg image.
-- While Conime can handle multiple images, only the first
three are displayed in viewports and only the first two
appear in the composite viewport.
IMAGE FORMAT
Conime currently accepts only 8 bit gray scale tiff images.
Typical images are approximately 5000x4000 pixels.
EXAMPLES
Enter:
conime master.cnm rlgs1.tif rlgs2.tif
to filter RLGS images rlgs1.tif and rlgs2.tif, register them
with a master profile, identify spots on rlgs1 and rlgs2
using spots on the master profile and create and save
rlgs1.cnm and rlgs2.cnm. Note that master.cnm will usually
not be in the current directory and so the path to
master.cnm will have to be specified.
After rlgs1.cnm and rlgs2.cnm are created, enter:
conime rlgs1.cnm rlgs2.cnm
to compare spots on rlgs1 and rlgs2 and mark mismatches.
ACKNOWLEDGEMENTS
Conime was developed in collaboration with Christoph Plass
and Dominic Smiraglia from the OSU Dept. of Human Cancer
Genetics. Funding was provided by The Ohio State University
Cancer Center and NIH.
AUTHORS
Tahimna Ansari, Jason Knight, Kazhiyur-Mannar Ramakrishnan,
Arunachalam Somasundarum, Rephael Wenger, Robert Wiryomar-
tini, Ronald Wiryomartini
Man(1) output converted with
man2html