[Tubetk-developers] TubeTK

[Stephen Aylward]

Hi,

1) converted your animated gif to a volume using ImageMagick and MetaIO.

2) Manually painted a few examples of small and large roots (see
PaintedRoots.png).

3) Created a training mask from the manually painted roots using
ImageMath (see mask.png)

> ImageMath all-label.mha -M 1 3 1 0 -W 2 d1.mha -M 1 3 1 0 -W 2 d2.mha -a 1 -1 d1.mha -a 127 255 all-label.mha -W 2 mask.mha

4) Used the mask to train a classifier to estimate the posterior
probability of being a root (see enhanced.png - left=before,
right=after - MIP views).   This trained classifier can then be
applied to other root images, without the need to re-train / use a
mask.

> EnhanceTubesUsingDiscriminantAnalysis --saveDiscriminantInfo roots.mdi --tubeScales 1,2,3,6,8,12 --labelmap mask.mha all.mha allEnh-t2.mha

We can also run the centerline traversal method to extract models of
the roots.  But first, do the enhanced results seem like what you
need?

Stephen

On Sun, Jun 29, 2014 at 3:40 PM, eli rosen <elirosen at gmail.com> wrote:
> Hi Stephen,
>
> I do have 3D data. It's an optical imaging technique, with light penetrating
> roots from one side. For overlapping roots, the root nearer to the optical
> source can shadow the farther root, meaning that the farther root will not
> have high contrast imaging for the shadowed region. Still, depth resolution
> allows roots that appear to be overlapping in one plane to be clearly
> distinguished.
>
> I've attached an animated gif, stepping through a representative 3D volume.
> Apologies for the size of the file. The actual data set is quite a bit
> larger.
>
> Best,
> Eli
>
>
> On Fri, Jun 27, 2014 at 10:17 AM, Stephen Aylward
> <stephen.aylward at kitware.com> wrote:
>>
>> Hi,
>>
>> If your data is in 3D, we might be able to help.
>>
>> The contrast seems to be generally good, but the crossing small roots
>> might be difficult to separate - the 3D resolution will probably be
>> more of a factor than the contrast.
>>
>> If you have a 3D volume, we could give it a try.
>>
>> Stephen
>>
>> On Tue, Jun 24, 2014 at 11:13 PM, eli rosen <elirosen at gmail.com> wrote:
>> > Thanks for your message, Stephen. It certainly looks like Tubetk could
>> > be
>> > interesting. I may attempt to start playing with it for my application,
>> > segmenting root networks of plants, via the 3D Slicer implementation.
>> > I've
>> > attached an image showing a representation of my data, and initial
>> > attempts
>> > at segmentation using Ilastik. For roots in soil, Ilastik was not able
>> > to
>> > discriminate between high contrast areas based on shape (I care about
>> > the
>> > roots but not the soil).
>> >
>> > Anyway, thanks again and I look forward to playing with this!
>> >
>> > Best,
>> > Eli
>> >
>> >
>> > On Mon, Jun 23, 2014 at 1:39 PM, Stephen Aylward
>> > <stephen.aylward at kitware.com> wrote:
>> >>
>> >> The methods are generally (mathematically) insensitive to intensity
>> >> variations.
>> >>
>> >> See the attached images for vessel enhancement --  Left: input.
>> >> Right: output.
>> >>
>> >> The extraction of centerlines is equally insensitive.
>> >>
>> >> s
>> >>
>> >> On Mon, Jun 23, 2014 at 1:35 PM, Ricardo Ortiz
>> >> <ricardo.ortiz at kitware.com> wrote:
>> >> > Hi Eli,
>> >> >
>> >> > It great that you are looking into TubeTK!
>> >> > I think the TubeTK folks should be able to answer that question with
>> >> > more
>> >> > expertise than I would.
>> >> > I am forwarding your email to the developers list.
>> >> >
>> >> > Cheers,
>> >> >
>> >> > /Ricardo
>> >> >
>> >> >
>> >> > On Mon, Jun 23, 2014 at 11:47 AM, eli rosen <elirosen at gmail.com>
>> >> > wrote:
>> >> >>
>> >> >> Hi Ricardo,
>> >> >>
>> >> >> I'm taking a peek at TubeTK now, and it looks really neat. Have you
>> >> >> evaluated its performance on any low contrast datasets? It looks
>> >> >> like
>> >> >> you
>> >> >> have some really nice contrast from the MRI and REUS data. Just
>> >> >> curious
>> >> >> how
>> >> >> well the segmentation works with lower contrast, which I have from
>> >> >> my
>> >> >> process.
>> >> >>
>> >> >> Thanks!
>> >> >>
>> >> >> -Eli
>> >> >>
>> >> >>
>> >> >> On Sat, Jun 21, 2014 at 11:35 AM, Ricardo Ortiz
>> >> >> <ricardo.ortiz at kitware.com> wrote:
>> >> >>>
>> >> >>> http://tubetk.org/
>> >> >>>
>> >> >>> /Ricardo
>> >> >>>
>> >> >>> Sent from phone, please excuse terseness and typos.
>> >> >>
>> >> >>
>> >> >
>> >> >
>> >> > _______________________________________________
>> >> > Tubetk-developers mailing list
>> >> > Tubetk-developers at public.kitware.com
>> >> > http://public.kitware.com/mailman/listinfo/tubetk-developers
>> >> >
>> >>
>> >>
>> >>
>> >> --
>> >> ====================================================
>> >> Stephen R. Aylward, Ph.D.
>> >> Senior Director of Operations, North Carolina, Kitware, Inc.
>> >> http://www.kitware.com and http://www.aylward.org
>> >> (919) 969-6990 x300
>> >> ====================================================
>> >
>> >
>>
>>
>>
>> --
>> ====================================================
>> Stephen R. Aylward, Ph.D.
>> Senior Director of Operations, North Carolina, Kitware, Inc.
>> http://www.kitware.com and http://www.aylward.org
>> (919) 969-6990 x300
>> ====================================================
>
>



-- 
====================================================
Stephen R. Aylward, Ph.D.
Senior Director of Operations, North Carolina, Kitware, Inc.
http://www.kitware.com and http://www.aylward.org
(919) 969-6990 x300
====================================================
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PaintedRoots.png
Type: image/png
Size: 165982 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/tubetk-developers/attachments/20140703/75ed816d/attachment-0003.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mask.png
Type: image/png
Size: 26400 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/tubetk-developers/attachments/20140703/75ed816d/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Enhanced.png
Type: image/png
Size: 521592 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/tubetk-developers/attachments/20140703/75ed816d/attachment-0005.png>