Posted by
Aryeh Weiss on
Nov 13, 2015; 11:04am
URL: http://imagej.273.s1.nabble.com/help-with-3D-segmentation-tp5014798p5014945.html
This last week I focused mostly on the 3D-watershed approach.
Updates are interleaved below.
On 08/11/2015 9:10 AM, Aryeh Weiss wrote:
Jan Eglinger posted the answer to this. Here is what he wrote:
-----
You have to go one step further: the `DifferenceOfGaussian` class has a
`findPeaks` method that gives you a list of `DifferenceOfGaussianPeak`s.
In its Javadoc [1] you find the methods:
boolean isMin()
boolean isMax()
that you can use to differentiate between minima and maxima.
I modified your script to only get the maxima:
https://gist.github.com/imagejan/7cc228b9a5e57cce44a0Here's what I changed:
https://gist.github.com/imagejan/7cc228b9a5e57cce44a0/revisions-----
This enables the threshold to be set lower, without picking up the minima.
> Among its parameters are Converter, OutOfBoundsStrategy (among others).
> I did not understand how to use these classes, so I wonder if someone on
> this list can tell me how to use them, or even better, point me at the
> documentation that might help (the class definitions in javadoc were not
> enough for me).
>
Jan also added the following:
-----
Albert's excellent tutorials unfortunately make use of the
now-deprecated ImgLib1 code that was superseded by ImgLib2 [2].
Jean-Yves Tinevez's TrackMate plugin is a great example how to use
ImgLib2. It's `LogDetector` [3] and `DogDetector` [4] classes show how
to detect spots using ImgLib2 implementations of LoG and DoG.
I added a script to the ImageJ wiki illustrating how to use
`DogDetector` from a Python script:
http://imagej.net/Jython_Scripting#Find_peaks_in_a_3D_image-----
I did not have a chance to try this, because I focused on the watershed
approach (see comments below).
> Thank in advance.
> --aryeh
>
>
> On 05/11/2015 12:58 PM, Aryeh Weiss wrote:
>> This is an update on the problem I posted.
>>
>> First, thanks to the people who replies with suggestions. From these, I
>> am now following up on two of them.
>>
>> 1. Albert Cardona suggested trying the 3D (actually multi-D) difference
>> of Gaussians (DoG) filter. This looks promising, plus it introduced me
>> to imglib. It works great on isolated cells (actually, everything works
>> great on isolated cells) and seems reasonable for my image, but still
>> needs work. because I have such a dense collection of nuclei, it
>> sometimes detects dark blobs, which are actually an approximately
>> cell-sized island within a sea of nuclei.
>> Albert's sample code displays the result in 3D using the 3D viewer.
>> Unfortunately, java3D does not work in java 1.8, and in Java 1.6, while
>> the java3D test works, the code that generates the 3D view fails -- that
>> will go into a separate thread.
>> The net result is that I can mark the centers of the nuclei (which is
>> why it look promising) but I cannot visualize it in 3D to see if it
>> really looks right.
>>
>> 2. Thomas Boudier suggest some 3D bandpass filtering followed by a
>> seeded watershed. I will work more on finding the right parameters for
>> this approach after I finish with the DoG approach. One problem is
>> that the resolution in Z is much lower that in X-Y. As a result, the
>> watershed does not separate the nuclei along Z. It is not clear to me
>> why the DoG does. However, I still need to learn to to use the output
>> of the watershed -- I may not be interpreting the output correctly.
>>
I found that I got a very good segmentation (ie, marking of the nuclei)
by using the watershed in the droplet finder plugins. Or more
accurately, using the droplet finder workflow which is run by
Plugins>Droplet Finder>DF_Filterstacker
The macro recorder enabled me to script it.
// The code to get the file and parse
// the filename and image title has been left out.
run("32-bit"); // DF-Bandpass wants 32 bit input
// The median, rolling ball filter, and Gaussian blur
// greatly improved the accuracy
run("Median...", "radius=2 stack");
run("Subtract Background...", "rolling=50 sliding stack");
run("Gaussian Blur...", "sigma=1.5 scaled stack");
// macro recorder created this. This is a long command, so if you
// cut and paste, you may have to reassemble it
run("DF_Filterstacker", "maximal_feature_size=15 minimal_feature_size=7
z/x_aspect_ratio=5.000 radius_x=2 radius_y=2 radius_z=2 invert
image=[BP_"+dupTitle+"] mask=[WS_BP_"+dupTitle+"] area_threshold=0.300
maximum_threshold=0.3 connect_threshold=0.80 slice=5");
IJ.renameResults("Results");
IJ.renameResults("Particle Results");
This script produces a table called "Particle Results", with (among
other things) the X, Y, and Z coordinates of the objects.
>> I recently got access to Imaris (basic) with a spot detector. It does a
>> decent job, but since I don't have it on my computer, I plan to solve
>> this with ImageJ.
>>
>> When I have something that works and that I can visualize, I will post
>> it for the benefit of other who, like me, have to muddle their way from
>> 2D to 3D.
>>
In order to visualize the results, I have a python script to read the
Particle Results table and produce a set of point ROIs that can mark the
original image.
The i can flatten the image and display this with the 3D Viewer (Java
1.6 only).
"""
This script converts the output of the droplet finder plugin (Paticle
Results)
to a set of point ROIs whcih can be used to mark the drops (in my case,
nuclei)
on the original image.
Input: Particle Results table from Droplet Finder
Output: An ROI Manager populated with point ROIs whose coordinates
correspond to the list in the table.
Bugs to fix: If the ROI Manager is open when the program is run,
rm.addRoi fails with a null pointer error.
"""
# Does not require auto-import
from ij import IJ, WindowManager
from ij.plugin.frame import RoiManager
from ij.measure import ResultsTable
from ij.gui import PointRoi
dfTitle = "Particle Results"
IJ.renameResults(dfTitle , "Results")
rt = ResultsTable.getResultsTable()
pos = []
for i in range(rt.size()):
pos.append([rt.getValue("position_x",i), rt.getValue("position_y", i),
rt.getValue("position_z", i)])
for i in range(rt.size()):
print pos[i];
print len(pos)
dfTitle = "Particle Results"
IJ.renameResults( "Results" , dfTitle )
# I do not know why this does not prevent the null pointer error
# which I get if the ROI Manager is open when this script is run
if "ROI Manager" in WindowManager.getImageTitles():
WindowManager.getWindow("ROI Manager").close()
rm = RoiManager()
rm.show()
for p in pos:
cellRoi = PointRoi(int(p[0]),int(p[1]))
cellRoi.setPosition(int(p[2]))
rm.addRoi(cellRoi)
rm.runCommand("Show All without labels")
# end of script
I actually created two images -- one is the original, and the second is
a blank image with the point ROIs flattened into it. Then I overlay them
with the 3D viewer and increase transparency of the data to see the
marks. What I really want to do is make the nuclei into surfaces and
then "peel" them away to reveal the marks. I have not yet figured that
out, and the surface rendering takes a really long time.
One problem I have is that sooner or later, the 3D Viewer menus get
grayed out (though mouse clicks in the 3D Viewer window still work).
The I have to close it and start again. I have not been able to
associate this with a reproducible set of actions.
I hope this is useful
--aryeh
--
Aryeh Weiss
Faculty of Engineering
Bar Ilan University
Ramat Gan 52900 Israel
Ph: 972-3-5317638
FAX: 972-3-7384051
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html