StepDeconvolution – Algorithm for Correction of
Data Extracted from Blurred Confocal Images
This work is
supported by NIH grant RR07801, GAP award RUB1-1578, NWO-RFBR project
047.011.2004.013, and RFBR grants 08-01-00315a and 08-04-00712a.
Description
StepDeconvolution is an algorithm
for the estimation and correction of errors in quantitative gene expression
levels obtained from blurred confocal images.
Quantification of confocal images is implemented by
means of image segmentation procedure that determines nuclear borders and
constructs the binary mask of an image. The mask is applied to a gene
expression image so that the information is only read from intranuclear
areas and intensity values are averaged over all the pixels composing the
nucleus. Even if the
mask is highly accurate, the blurring of an image due to diffraction scattering
in confocal microscope distorts the intensity values at sharp nuclear borders.
This results in underestimation of the mean levels of fluorescence intensity
and hence gives rise to inaccuracy in the quantitative data. The direct deconvolution
of images by the standard Richardson-Lucy method [1] does not eliminate the
blurring at the nucleus borders and provides only a partial correction of mean
intensities. StepDeconvolution is a modification
of the RL method that provides more precise restoration of the data which is read
from blurred images of objects with sharp edges. The detailed description of the method is
given in [2].
1.
van Kempen,G.M.P. and van Vliet,L.J.,
Background estimation in non linear image restoration, Journal of Optical Society of America, 2000, v.A17,
425-433.
2.
E.Myasnikova, S.Surkova, L.Panok, M.Samsonova and J.Reinitz Estimation
of errors in gene expression data introduced by confocal imaging, submitted to Bioinformatics
Description of input files and parameters
Input files
are
- Image of an embryo stained for
gene expression in tiff format (see sample file HETaa12.tif)
- Text file containing
information about the binary mask image (see sample file HETaa12.blob)
Format of the file:
<blob>
# of nuclei(N) # of rows #
of columns
only_nuclei
N-1
< x-coordinate of (N-1)th
nucleus(% of embryo length)>
<y-coordinate of (N-1)th nucleus (% of
embryo width)> <# of pixels
belonging to (N-1)th nucleus>
list of pixels belonging to (N-1)th nucleus
N-2 <
x-coordinate of (N-2)th
nucleus(% of embryo length)>
<y-coordinate of (N-2)th nucleus (% of
embryo width)> <# of pixels
belonging to (N-2)th nucleus>
list of pixels belonging to (N-2)th nucleus
…..
0
< x-coordinate of 0th nucleus(% of embryo
length)> <y-coordinate of 0th
nucleus (% of embryo width)> <#
of pixels belonging to 0th nucleus>
list of pixels belonging to 0th nucleus
</blob>
- input parameter <pixel size
(mkm)>
- size of pixel in the input image (μm)
- input parameter <initial
value of blur parameter> (optional) – initial value of blur parameter,
default value is 1.5μm-1
Description of output files and parameter estimates
Output
files are
1. Result file <output_file>
containing raw and corrected gene expression data (see sample file HETaa12_h).
Format of the file:
n x y mean corrected mean variance corrected
variance step error npixel
n #of
a nucleus
x x-coord
y y-coord
mean
raw mean intensity
corrected mean corrected mean intensity
variance raw
within-nucleus variance
corrected variance corrected w-n variance
step estimated
step image
error difference
between corrected and raw means
npixel #
of pixels in the nucleus
2. row0_< output_file>
work file illustrating the method. It presents a central row of pixels
extracted from the input image along x-axis (see sample file row0_HETaa12_h).
Format of the file:
n blurred_step raw step
n #of
a nucleus
blurred_step estimated blurred step image
raw observed
input image
step estimated
step image (pixels excluded from fitting are shown zero)
3. row_ <output_file> work file illustrating the method.
It presents a central row of pixels extracted from the input image along x-axis (see sample file
row_HETaa12_h).
4.
Format of the file:
n step restored raw blurred restored residuals
n #of
a nucleus
step estimated
step image
restored restored
image
raw observed
input image
blurred_restored blurred
restored image
residuals difference
between the observed and blurred step images to be restored
5. Estimate of the blur parameter is
output to the screen
Download StepDeconvolution (tgz)
Download sample files (tgz)
Installation
Download StepDeconvolution
and unpack the archive in your home directory.
Running
Download
sample files archive and unpack it.
Run the StepDeconvolution from the terminal
with the command:
[user@machine ~]$ ./ step_deconvolution_fc9_x86 <input gene expression image in tiff
format> <text file with information about the mask> <output
file> <pixel size(mkm)> <initial value of
blur parameter(optional)>
Example
[user@machine ~]$ ./
step_deconvolution_fc9_x86 HETaa12.tif
HETaa12.blob HETaa_12_h 4.8
The tool should run on x86
computer with recent GNU/Linux distro - Fedora 9, Ubuntu
8.04
Written in C-language, compiled
on Fedora 9 x86 by gcc 4.3 compiler. May run on other systems.
