Supplemental material for:

taxidermistplateSoftware and s/w Development

Nov 7, 2013 (3 years and 8 months ago)

96 views


Supplemental material

for:


Improved structure, function, and compatibility for CellProfiler:

modular high
-
throughput image analysis software


Lee Kamentsky
1
, Thouis R. Jones
1
, Adam Fraser
1
, Mark
-
Anthony Bray
1
, David J. Logan
1
,
Katherine L. Madden
1
, Vebjo
rn Ljosa
1
, Curtis Rueden
2
, Kevin W. Eliceiri
2
, and Anne E.
Carpenter
1
*


1 Imaging Platform, Broad Institute of Harvard and MIT, Cambridge, Massachusetts, USA 02142

2 Laboratory for Optical and Computational Instrumentation, Univ. Wisconsin, Madison,
Wiscon
sin, USA 53706

*To whom correspondence should be addressed.

1. Comparison of CellProfiler 2.0
versus

1.0


The figure and tables below provide an overview of the
similarities and differences in
features between
CellProfiler 2.0 and its previous version, Ce
llProfiler 1.0.




Supplemental

Figure 1: Enhanced usability and features in CellProfiler 2.0.

(A) CellProfiler 2.0 can run ImageJ plugins. Here, the results of running an Ima
geJ plug
-
in, Tubeness,
within a CellProfiler pipeline are shown (contrast
-
enhanced,
http://www.longair.net/edinburgh/imagej/tubeness/).

(B) CellProfiler 2.0 monitors users’ selections and immediately flags problems. Here, an error in the
p楰敬楮攠楳⁢ioug
ht to the user’s attention.

(C) CellProfiler 2.0’s new test mode allows stepping through a pipeline and adjusting the settings for
敡捨emoTu汥⸠

⡄⤠H敬e⁢ 瑴on猠s硩獴sfo爠e慣a⁳ W瑩Wg⁩渠敡捨 moTu汥l楮⁃ 汬Pro晩f敲e2⸰ⰠHo⁴he⁵ 敲e捡c⁩mm敤e慴敬e
ob瑡楮

gu楤慮捥c睨敮⁣woo獩sg⁳ 瑴楮g猠楮⁡⁰sp敬楮攮



Supplemental Table 1:
Similarities between CellProfiler 2.0 and CellProfiler 1.0


-

Open
-
source code (GPL v.2)


-

Flexible set of modules for image analysis


-

Interface enables modules to be linked together to

create image
analysis pipelines, which can be saved, reused, and shared


-

Can be run with an interactive GUI, or headless and on a computing
cluster


-

Variety of output options, including spreadsheets and databases


-

Compatible with CellProfiler Analyst d
ata exploration software


-

Usable by researchers who lack computational skills, extensible by
computer scientists developing new algorithms


-

Can read CellProfiler 1.0 pipeline file format


Supplemental Table 2:
Improvements in CellProfiler 2.0
versus

Cel
lProfiler 1.0



Infrastructure and
interoperability

-

Open
-
source Python programming language (CellProfiler
1.0 used the proprietary Matlab programming language,
which required a license for developing new modules)


-

Speed (see
Supplemental Figure 2
)


-

Can

run ImageJ plugins


-

Uses BioFormats, which expands the number of readable
and writable image file formats


-

More convenient to set up CellProfiler Analyst to explore
CellProfiler 2.0
-
produced data


-

Comprehensive unit testing and nightly builds (Windows
and Mac OSX)


-

New developer tools: developer's guide (wiki), email lists,
open SVN repository, and code browsing tools


-

Automatic crash reporting


-

Plugin system supports custom modules, even in the
distributed version



Usability

-

Test mode enables a c
onvenient interface for stepping
through a pipeline to debug it and check results


-

Continual verification
-

immediate feedback is given when
a pipeline error is detected


-

Context
-
dependent module settings
-

only the settings that
are relevant to the exis
ting settings choices are shown


-

Ability to undo recent edits to a pipeline


-

Many modules and settings have been consolidated for
simplicity


-

Modules can be copied, pasted, dragged, and dropped
when creating pipelines


-

Module windows' show/hide status
can be saved


-

Help has been completely revised


-

Help is available for each individual setting in each module

(continued, next
page)




Algorithms/

features

-

New three
-
class thresholding algorithm for images that
have regions of intermediate intensity in

addition to
foreground and background


-

Improved measurement of Zernike shape features and
Haralick

texture features (
Supplemental Figure 3
)


-

New UnmixColors module performs color deconvolution for
histologically
-
stained images


-

Objects can be filtered
on the basis of any number of
different measurements, including machine
-
learning
-
based
classifiers


-

TrackObjects module contains the new LAP algorithm
(Jaqaman, et al., Nature Methods 2008), capable of
bridging temporal gaps in trajectories and accountin
g for
splitting/merging events


-

EnhanceOrSuppressFeatures module improves the
detection of neurites and dark spots surrounded by bright
rings


-

New MeasureNeurons module measures the number of
trunks and branches for each neuron in an image


-

CorrectIllum
inationCalculate module has a new spline
method for determining the background intensity across an
image


-

MeasureGranularity and Mask can now function on objects
rather than just images


-

Convex hull is now available as an image morphological
operation an
d as a background correction method



High
-
throughput
enhancements

-

Data can be uploaded directly to a database during a
processing run, allowing data exploration and analysis to
begin while some jobs are still processing


-

Images can be analyzed based on

a comma
-
delimited text
file of image file locations (which could be exported by a
microscope database or laboratory information
management system)


-

Images can be loaded via HTTP and FTP


-

Unlimited amounts of text information about the images
(metadata)
can be loaded via a comma
-
delimited file


-

Images and resulting data can be grouped for aggregate
operations based on metadata (e.g., to correct illumination
variation on a per
-
plate basis, calculate average
phenotypic readout on a per
-
well basis)


-

Data c
an be exported in SQLite, a self
-
contained database
format, to enable creation of a local database. As
CellProfiler Analyst also supports SQLite, its data
exploration and machine
-
learning tools can be used on
CellProfiler
-
produced data without
the
need
for

a separate
database server.


-

FlagImage module can exclude images from analysis based
on image quality


-

Object measurement data can be exported to a database
as a single table or as separate tables, one for each object


-

M
etadata substitution in settings

is available

via

a
metadata editor

(e.g.
,
custom

file names in SaveImages
)

2. Analysis of CellProfiler speed


Goal:

We compared the speed of CellProfiler 2.0 (CP2.0) to that of CellProfiler 1.0 (CP1.0) using a typical
image analysis pipeline.

Methods:
C
ellProfiler is optimized for batch execution on a single CPU core. An experimental analysis is
typically broken into one batch per available CPU core and each batch is run within a separate
CellProfiler instance. Therefore, we benchmarked the CPU time requ
ired to run CellProfiler instead of
the wall
-
clock time because the CPU time is the limiting resource. CP2.0 and CP1.0 were run five times
on the same computer: a 2
-
CPU
Quad
-
Cor
e AMD Opteron


processor (2300 MHz, 8 cores total) with
16 Gb of memory runni
ng Linux 2.6 / Red Hat 4.1. Some of the performance variability was due to this
cluster node being a shared resource, but the median times were judged to be representative.

We used a set of Society for Biomolecular Screening example images
(
http://www.cellprofiler.org/examples.shtml

Human cytoplasm
-
nucleus translocation assay) to
benchmark speed. We made two changes to the example CP1.0 and CP2.0 pipelines to make them
equivalent: we enabled Zernik
e feature extraction in the CP1.0 pipeline to match the CP2.0 pipeline and
we measured two angles instead of four for the Gabor features in the CP2.0 pipeline to match the
functionality of CP1.0 (CP1.0 calculates the Gabor filter in the X and Y direction a
nd is not configurable).
During the course of our investigation, we discovered an opportunity to optimize the performance of
CP2.0’s Mixture of Gaussian background method without compromising its accuracy. The performance
measurements for CP2.0 have been m
ade on the optimized code, SVN revision 10843 of CellProfiler 2.0
and CP1.0. The SBS images consist of 96 image sets. Each image set is composed of a 640 x 640 pixel
single
-
channel image of a nuclear stain and a corresponding single
-
channel image of a Fork
head / GFP
fusion protein which translocates to the nucleus when treated with a positive control perturbation.

Results:
Overall, CP2.0 was faster than CP1.0. For the pipeline and images described above, the median
CPU time per image set analyzed was 51 CP
U
-
seconds (standard deviation of 4 seconds) for CP2.0 versus
74 CPU
-
seconds (standard deviation of 19 seconds) for CP1.0 (
Supplemental Figure 2
). Excluding the
two most time
-
consuming modules, MeasureTexture and MeasureObjectSizeShape (described below),
CP
2.0 is slightly faster than CP1.0 (22.7 CPU seconds / image set for CP2.0, 25.9 CPU seconds / image set
for CP1.0). This reflects a general speed improvement attributable to algorithm optimization and
vectorization and improvements attributable to the perf
ormance of Python’s Num
P
y numerical
processing library.

The major improvement in processing time is due to changes in the calculation of Zernike features in the
MeasureObjectSizeShape module (45 seconds / image set in CP1.0, 18 seconds / image set in CP2.0
).
The Zernike calculation is one
optimization
among many during the port to Python

achieved

using
vectorization and other techniques. Vectorization reduces looping operations performed on single
quantities to single operations performed on vectors. The re
sulting code uses a processor’s superscalar
operations and
makes better use of

the processor cache. In addition, the CPU time used in a vectorized
operation scales linearly with image size or with the number of pixels in an image’s objects which
accounts f
or the lower variance in CPU time per image set analyzed for CP2.0.


The only notable decrease in CP2.0’s performance was in the MeasureTexture module (10 seconds /
image set in CP2.0, 2.5 seconds / image set in CP1.0), but this is due to increased funct
ionality. In both
CP 1.0 and 2.0, MeasureTexture calculates two classes of features: Haralick
i

features and features based
on a Gabor
-
filtered image
ii

iii
. The Gabor
-
based features respond to striped texture. CP1.0 constructs a
single kernel for the Gabor fil
ter operation. The kernel has a fixed size of slightly less than the median
radius of the objects in an image and a single exponential fall
-
off based on this median radius. The
texture of pixels not covered by the kernel is not measured. Instead, CP2.0 per
forms a vectorized
calculation of the Gabor filter, properly scaled to the size of the object being measured and covering all
pixels in the object
(
Supplemental Figure
3
)
.
Furthermore, CP2.0’s Gabor filter can be calculated at a
user
-
selected number of ang
les whereas CP1.0’s Gabor filter is calculate
d only at angles of 0° and 90°.

Supplemental Figure 2: Comparison of CellProfiler 2.0 and 1.0 in terms of speed to run a typical
pipeline.

See text for details.





3. Analysis of improvements to the Haralick features


We improved the measurement of Haralick texture features in CellProfiler 2.0. The basic al
gorithm for
calculating Haralick features quantizes the intensity range in the region of interest into a discrete
number of bins. The algorithm then calculates statistics based on the correlation of the quantized values
for nearby pixels. CP1.0 calculates
a single quantization scale per image based on the minimum and
maximum pixel values in that image. CP2.0 improves on this
approach
by calculating separate
quantization scales per measured object; each quantization uses the maximum dynamic range of
intensit
ies available for the object. This yields more information and avoids inadvertently measuring
intensity instead of texture.

To test whether this adjustment improves the ability to measure meaningful properties of cells, we
measured the ability of the Haral
ick texture features in CP1.0 and CP2.0 to distinguish positive and
negative controls in a texture
-
based assay. Images were taken from the Society for Biomedical
Screening CompuCyte Transfluor image set (available at
http://www.broadinstitute.org/bbbc/sbs_compucyte_transfluor.html
). This image set consists of images
of cells treated with twelve different doses of
isoproterenol

which, at higher doses, results in a speckled
appe
arance in the GFP image. Cells in the image were first identified using CellProfiler 2.0 so that the
Haralick features could be measured using identical cell boundaries for both CP1.0 and CP2.0. CP1.0 and
CP2.0 were then run with pipelines that measured t
he Haralick features in the GFP channel at scales of
Supplemental Figure 3: The Gabor filter in CellProfiler 2.0 (left)

differs
from that of CellProfiler 1.0
(right).
See text for details.

3, 6 and 10 pixels.
We calculated the Z’ factor
iv

(a measure of assay quality) based on the replicates at
highest and lowest doses of
isoproterenol

using CellProfiler’s CalculateStatistics module (
Supplem
ental
Table 3
).

CP2.0 yielded the best Z’ factor for the assay (0.53)
versus

CP1.0’s best Z’ factor (0.49). In fact, all
Haralick features were more informative in CP2.0
versus

CP1.0, as measured by the Z
-
factor, with one
exception: the least informative f
eature overall
(
Difference

Variance)
had a higher Z
-
factor as measured
by CP1.0.


Supplemental Table 3: Z’ factors of Haralick features as calculated by CellProfiler 1.0 and 2.0.
The
features are listed in order of ability to distinguish positive and nega
tive controls (higher Z’ factor is
better).

Haralick feature

Scale in pixels


CP2
.0


CP1
.0

Angular Second Moment

10


0.52605987


0.488999

Sum

Average

3


0.52509592


0.408688

Angular

Second

Moment

6


0.52431513


0.471771

Sum

Average

6


0.51367252


0.390076

Angular

Second

Moment

3


0.50207589


0.426094

Sum

Average

10


0.49279165


0.369324

Sum

Variance

10


0.43333049


0.369852

Variance

10


0.42922917


0.389806

Sum

Variance

6


0.41699802


0.333137

Variance

6


0.40331845


0.343607

Su
m

Variance

3


0.39721828


0.294448

Variance

3


0.39476942


0.275976

Entropy

3


0.39346471


0.287969

Correlation

6


0.35879887


0.186271

InfoMeas2

6


0.35688185


0.185321

Entropy

6


0.34191769


0.217906

Entropy

10


0.32826782


0.211466

Su
m Entropy

3


0.32504093


0.189567

Sum Entropy

6


0.30605741


0.150692

Sum Entropy

10


0.29321212


0.141536

Contrast

10


0.24822979


0.176758

InfoMeas2

10


0.23282403


0.083112

Contrast

6


0.15755930


0.043928

Difference

Entropy

3


0.15503972


0.005034


Inverse

Difference

Moment

10


0.15405655


-
0.117193

Correlation

10


0.11616801


-
0.000134

Haralick feature

Scale in pixels


CP2
.0


CP1
.0

InfoMeas2

3


0.10148035


-
0.146025

InfoMeas1

6


0.07360163


-
0.110223

Correlation

3


0.05897995


-
0.217220

Difference

Entropy

10


0.036153
58


-
0.143419

Difference

Entropy

6


-
0.02216420


-
0.244628

Difference

Variance

10


-
0.13290311


-
0.159667

Inverse

Difference

Moment

6


-
0.16151087


-
0.587989

InfoMeas1

10


-
0.16570178


-
0.301811

InfoMeas1

3


-
0.24827404


-
0.487593

Inverse

Difference

Moment

3


-
0.49739798


-
0.996050

Difference

Variance

6


-
1.39976935


-
1.533185

Contrast

3


-
1.87827603


-
2.332562

Difference

Variance

3


-
8.39944329


-
8.099111


4. Supplemental

Data: Example CellProfiler 2.0 pipeline file


The pipel
ine below is presented as plain text. An updated version of this pipeline, “ExampleFly.cp”, is a
text file that can be downloaded from http://www.cellprofiler.org/examples.shtml


----------------------------------------------------------------------------
-------------------------------------------------


CellProfiler Pipeline: http://www.cellprofiler.org


Version:1


SVNRevision:9722




LoadImages:[module_num:1|svn_version:
\
'9660
\
'|variable_revision_number:5|show_window:True|no
tes:
\
x5B
\
x5D]



File type t
o be loaded:individual images



File selection method:Text
-
Exact match



Number of images in each group?:3



Type the text that the excluded images have in common:Do not use



Analyze all subfolders within the selected folder?:No



Input ima
ge file location:Default Input Folder
\
x7C.



Check image sets for missing or duplicate files?:No



Group images by metadata?:No



Exclude certain files?:No



Specify metadata fields to group by:



Text that these images have in common (case
-
sensitive):D.TIF



Name this loaded image:OrigBlue



Position of this image in each group:D.TIF



Extract metadata from where?:None



Regular expression that finds metadata in the file name:None



Type the regular expression that finds metad
ata in the subfolder path:None



Text that these images have in common (case
-
sensitive):F.T



Name this loaded image:OrigGreen



Position of this image in each group:F.T



Extract metadata from where?:None



Regular expression that finds met
adata in the file name:None



Type the regular expression that finds metadata in the subfolder path:None



Text that these images have in common (case
-
sensitive):R.T



Name this loaded image:OrigRed



Position of this image in each group:R.T




Extract metadata from where?:None



Regular expression that finds metadata in the file name:None



Type the regular expression that finds metadata in the subfolder path:None




Crop:[module_num:2|svn_version:
\
'9633
\
'|variable_revision_number:2|sho
w_window:True|notes:
\
x5
B
\
x5D]



Select the input image:OrigBlue



Name the output image:CropBlue



Select the cropping shape:Rectangle



Select the cropping method:Coordinates



Apply which cycle
\
's cropping pattern?:First



Left and righ
t rectangle positions:501,700



Top and bottom rectangle positions:251,450



Coordinates of ellipse center:200,500



Ellipse radius, X direction:400



Ellipse radius, Y direction:200



Use Plate Fix?:No



Remove empty rows and columns?:Ed
ges



Select the masking image:None



Select the image with a cropping mask:None



Select the objects:None




Crop:[module_num:3|svn_version:
\
'9633
\
'|variable_revision_number:2|show_window:True|notes:
\
x5
B
\
x5D]



Select the input image:OrigGreen



Name the output image:CropGreen



Select the cropping shape:Previous cropping



Select the cropping method:Coordinates



Apply which cycle
\
's cropping pattern?:First



Left and right rectangle positions:300,600



Top and bottom rectang
le positions:300,600



Coordinates of ellipse center:500,500



Ellipse radius, X direction:400



Ellipse radius, Y direction:200



Use Plate Fix?:No



Remove empty rows and columns?:Edges



Select the masking image:None



Select the im
age with a cropping mask:CropBlue



Select the objects:None




Crop:[module_num:4|svn_version:
\
'9633
\
'|variable_revision_number:2|show_window:True|notes:
\
x5
B
\
x5D]



Select the input image:OrigRed



Name the output image:CropRed



Select the cro
pping shape:Previous cropping



Select the cropping method:Coordinates



Apply which cycle
\
's cropping pattern?:First



Left and right rectangle positions:300,600



Top and bottom rectangle positions:300,600



Coordinates of ellipse center:5
00,500



Ellipse radius, X direction:400



Ellipse radius, Y direction:200



Use Plate Fix?:No



Remove empty rows and columns?:Edges



Select the masking image:None



Select the image with a cropping mask:CropBlue



Select the objects
:None




IdentifyPrimaryObjects:[module_num:5|svn_version:
\
'9633
\
'|variable_revision_number:6|show_windo
w:True|notes:
\
x5B
\
x5D]



Select the input image:CropBlue



Name the primary objects to be identified:Nuclei



Typical diameter of objects, in p
ixel units (Min,Max):10,40



Discard objects outside the diameter range?:Yes



Try to merge too small objects with nearby larger objects?:No



Discard objects touching the border of the image?:Yes



Select the thresholding method:MoG Global




Threshold correction factor:1.6



Lower and upper bounds on threshold:0,1



Approximate fraction of image covered by objects?:0.2



Method to distinguish clumped objects:Intensity



Method to draw dividing lines between clumped objects:Intens
ity



Size of smoothing filter:10



Suppress local maxima that are closer than this minimum allowed distance:5



Speed up by using lower
-
resolution image to find local maxima?:Yes



Name the outline image:None



Fill holes in identified obje
cts?:Yes



Automatically calculate size of smoothing filter?:Yes



Automatically calculate minimum allowed distance between local maxima?:Yes



Manual threshold:0.0



Select binary image:MoG Global



Retain outlines of the identified objects
?:No



Automatically calculate the threshold using the Otsu method?:Yes



Enter Laplacian of Gaussian threshold:.5



Two
-
class or three
-
class thresholding?:Two classes



Minimize the weighted variance or the entropy?:Weighted variance



Assi
gn pixels in the middle intensity class to the foreground or the background?:Foreground



Automatically calculate the size of objects for the Laplacian of Gaussian filter?:Yes



Enter LoG filter diameter:5



Handling of objects if excessive number

of objects identified:Continue



Maximum number of objects:500




IdentifySecondaryObjects:[module_num:6|svn_version:
\
'9633
\
'|variable_revision_number:4|show_win
dow:True|notes:
\
x5B
\
x5D]



Select the input objects:Nuclei



Name the objects to be i
dentified:Cells



Select the method to identify the secondary objects:Propagation



Select the input image:CropGreen



Select the thresholding method:Otsu Global



Threshold correction factor:1



Lower and upper bounds on threshold:0,1



Approximate fraction of image covered by objects?:10%



Number of pixels by which to expand the primary objects:10



Regularization factor:0.05



Name the outline image:Do not use



Manual threshold:0



Select binary image:Do not use



Re
tain outlines of the identified secondary objects?:No



Two
-
class or three
-
class thresholding?:Two classes



Minimize the weighted variance or the entropy?:Weighted variance



Assign pixels in the middle intensity class to the foreground or the ba
ckground?:Foreground



Discard secondary objects that touch the edge of the image?:No



Discard the associated primary objects?:No



Name the new primary objects:FilteredNuclei



Retain outlines of the new primary objects?:No



Name the new
primary object outlines:FilteredNucleiOutlines




IdentifyTertiaryObjects:[module_num:7|svn_version:
\
'9633
\
'|variable_revision_number:1|show_windo
w:True|notes:
\
x5B
\
x5D]



Select the larger identified objects:Cells



Select the smaller identified obje
cts:Nuclei



Name the tertiary objects to be identified:Cytoplasm



Name the outline image:Do not use



Retain outlines of the tertiary objects?:No




MeasureObjectSizeShape:[module_num:8|svn_version:
\
'1
\
'|variable_revision_number:1|show_windo
w:Tr
ue|notes:
\
x5B
\
x5D]



Select objects to measure:Cells



Select objects to measure:Nuclei



Select objects to measure:Cytoplasm



Calculate the Zernike features?:No




MeasureObjectIntensity:[module_num:9|svn_version:
\
'9660
\
'|variable_revision_nu
mber:3|show_wind
ow:True|notes:
\
x5B
\
x5D]



Hidden:1



Select an image to measure:CropBlue



Select objects to measure:Nuclei



Select objects to measure:Cells



Select objects to measure:Cytoplasm




MeasureTexture:[module_num:10|svn_version:
\
'1
\
'|variable_revision_number:1|show_window:True|
notes:
\
x5B
\
x5D]



Hidden:1



Hidden:3



Hidden:1



Select an image to measure:CropBlue



Select objects to measure:Nuclei



Select objects to measure:Cells



Select objects to measure:C
ytoplasm



Texture scale to measure:3



Number of angles to compute for Gabor:4




GrayToColor:[module_num:11|svn_version:
\
'9633
\
'|variable_revision_number:2|show_window:True|
notes:
\
x5B
\
x5D]



Select a color scheme:RGB



Select the input image
to be colored red:CropRed



Select the input image to be colored green:CropGreen



Select the input image to be colored blue:CropBlue



Name the output image:RGBImage



Relative weight for the red image:1



Relative weight for the green imag
e:1



Relative weight for the blue image:1



Select the input image to be colored cyan:None



Select the input image to be colored magenta:None



Select the input image to be colored yellow:None



Select the input image that determines brigh
tness:None



Relative weight for the cyan image:1



Relative weight for the magenta image:1



Relative weight for the yellow image:1



Relative weight for the brightness image:1



Select the input image to add to the stacked image:None




Sa
veImages:[module_num:12|svn_version:
\
'9679
\
'|variable_revision_number:5|show_window:True|n
otes:
\
x5B
\
x5D]



Select the type of image to save:Image



Select the image to save:RGBImage



Select the module display window to save:RGBImage



Select m
ethod for constructing file names:From image filename



Select image name for file prefix:OrigBlue



Enter single file name:OrigBlue



Do you want to add a suffix to the image file name?:Yes



Text to append to the image name:RGB



Select fi
le format to use:tif



Output file location:Default Output Folder
\
x7CNone



Image bit depth:8



Overwrite existing files without warning?:No



Select how often to save:Every cycle



Rescale the images? :No



Select colormap:gray



Upda
te file names within CellProfiler?:No



Create subfolders in the output folder?:No




ExportToSpreadsheet:[module_num:13|svn_version:
\
'9660
\
'|variable_revision_number:6|show_windo
w:True|notes:
\
x5B
\
x5D]



Select or enter the column delimiter:Comma (",
")



Prepend the output file name to the data file names?:Yes



Add image metadata columns to your object data file?:No



Limit output to a size that is allowed in Excel?:No



Select the columns of measurements to export?:No



Calculate the
per
-
image mean values for object measurements?:Yes



Calculate the per
-
image median values for object measurements?:No



Calculate the per
-
image standard deviation values for object measurements?:No



Output file location:Default Output Folder
\
x7C
.



Export all measurements?:No



Press button to select measurements to export:None
\
x7CNone



Data to export:Image



Combine these object measurements with those of the previous object?:No



File name:Image.csv



Use the object name for
the file name?:No



Data to export:Nuclei



Combine these object measurements with those of the previous object?:No



File name:Nuclei.csv



Use the object name for the file name?:No



Data to export:Cells



Combine these object measureme
nts with those of the previous object?:No



File name:Cells.csv



Use the object name for the file name?:No



Data to export:Cytoplasm



Combine these object measurements with those of the previous object?:No



File name:Cytoplasm.csv



U
se the object name for the file name?:No


-----------------------------------------------------------------------------------------------------------------------------


5. References




i

R Haralick & Its’hak Dinstein,
Textural Features for Imag
e Classification
, IEEE Transactions on Systems, Man and
Cybernetics Vol SMC
-
3, 1973, pp 610
-
621

ii

J.G. Daugman,

Uncertainty relations for resolution in space, spatial frequency, and orientation optimized by two
-
dimensional visual cortical filters
, Journal
of the Optical Society of America A, 1985, vol. 2, pp. 1160
-
1169

iii

Gabor, D. ,
Theory of communication

Journal of the Institute

of Electrical Engineers, vol 93, (1946) pp 429
-
441

iv

J Zhang,
A Simple Statistical Parameter for Use in Evaluation and Validatio
n of High Throughput Screening Assays
,
Journal of Biomedical Screening, Vol. 4, (1999) pp 67
-
73