3D Watershed Plugin?

[Zwick, Ari A.]

Dear ImageJ Community,

 

This is a continuation of a previous question, whereby I need one final
push to the finish.  I have multiple stacks of grayscale images that
require 3D watershed segmentation in order for me to obtain volume and
radii of pore space in soil aggregates.  Michael Doube and Robert
Dougherty were very helpful in providing their plugins: local thickness,
and BoneJ.  I have since decided to go with the particle analyzer in
BoneJ, but the final requirement is that I separate pores in 3D before I
use the particle analyzer.  It turns out that many of my pores are
connected and it becomes impossible to obtain the feret diameter with
such large networks.  2D watershed doesn't work because it thresholds at
different spots in different slices.  I tried using the IFT-3D plugin in
Fiji, which works on grayscale images, but it hasn't given me the
results I'm looking for and I can't find any documentation on it.  I was
able to find a 3D RGB watershed plugin, but that doesn't do me any good
with grayscale images.  Does anyone know of a 3D watershed plugin that
works on greyscales, or have any other ideas of how I could get around
this?

 

One more alternative I thought of was finding a Maximal Balls algorithm,
for which even if it just made spheres I could still use the particle
analyzer to get the radius.  I haven't seen any ImageJ plugins that do
that though, has anyone else?

 

Thanks in advance, I really appreciate any and all help that I've had in
the past, and hopefully in the future.

 

Ari Zwick

 

[Michael Schmid]

Hi Ari,

if you have a 3D Euclidian Distance Map (EDM), you can simply search  
for the local maxima of this EDM. They will be the centers of the  
largest inscribed spheres, with the value equal to the radius.
There might be multiple nearby maxima (e.g. imagine a tube where you  
can inscribe many spheres with similar radius). You will probably  
need some type of post-processing to get rid of this.

Michael
________________________________________________________________

On 6 Apr 2010, at 00:47, Zwick, Ari A. wrote:

> Dear ImageJ Community,
>
>
>
> This is a continuation of a previous question, whereby I need one  
> final
> push to the finish.  I have multiple stacks of grayscale images that
> require 3D watershed segmentation in order for me to obtain volume and
> radii of pore space in soil aggregates.  Michael Doube and Robert
> Dougherty were very helpful in providing their plugins: local  
> thickness,
> and BoneJ.  I have since decided to go with the particle analyzer in
> BoneJ, but the final requirement is that I separate pores in 3D  
> before I
> use the particle analyzer.  It turns out that many of my pores are
> connected and it becomes impossible to obtain the feret diameter with
> such large networks.  2D watershed doesn't work because it  
> thresholds at
> different spots in different slices.  I tried using the IFT-3D  
> plugin in
> Fiji, which works on grayscale images, but it hasn't given me the
> results I'm looking for and I can't find any documentation on it.  
> I was
> able to find a 3D RGB watershed plugin, but that doesn't do me any  
> good
> with grayscale images.  Does anyone know of a 3D watershed plugin that
> works on greyscales, or have any other ideas of how I could get around
> this?
>
>
>
> One more alternative I thought of was finding a Maximal Balls  
> algorithm,
> for which even if it just made spheres I could still use the particle
> analyzer to get the radius.  I haven't seen any ImageJ plugins that do
> that though, has anyone else?
>
>
>
> Thanks in advance, I really appreciate any and all help that I've  
> had in
> the past, and hopefully in the future.
>
>
>
> Ari Zwick
>
>

[Teresa W]

Hi Avi

I'm interested in this as well, as I want to quantify the void size distributions for my highly interconnected porous bread samples. Please, if you could pass on what you decided to do regarding 3d watershedding, that would be very helpful.

If you haven't found a solution, I worked out a way to watershed particles/voids in stacks for 2D counting. The approach arbitarily cuts off intersecting voids at necks close to the edge of the counting frame. As I wanted to get an unbiased count of particle size categories, I used different counting frames to exclude particles intersecting two sides of the final count frame, while stitching back the spillover of particles on the other two sides. See Disector Counting Frame post Jun 11, 2008 for the details, my post's towards the end.

I also wanted to count the particles with watershedding removed (Coaleseced/interconnected). For this
* Subtracted watershedded whole image from original binary image to get watershed lines
* ANDed the lines with a mask of the counted particles which had undergone Erode/Dilate,
* Added back into stack to remove watershed lines from my binary counting image.
* Recounted reconnected image.

I think this approach would work in 3D as well
* Follow process in 2D
* Check stack, flicking through to erase the occasional extra particle that appears on the edges.

I think it would be useful also to quantify the number of 'necks' intersecting the count volume where the voids were cut off, as part of defining the counted unit within the sample.

Anyway, I am going to give it a go myself now.

Teresa Wegrzyn
Food Tecnology
Massey University
Palmerston North
New Zealand