Automated segmentation and analysis of Arabidopsis seedlings

[Andrew Defries]

Dear Imagej and Fiji users,

 I have an image analysis problems and would like your ideas and opinions.
I will present here, briefly, my image processing goals to get us started:

 My images are large so I have been working on a cropped version to prevent
crashes, but downstream dealing with whole 4000 x 3000 may be a problem.
I'm not concerned now, but if we can simplify my analysis/process to make
larger photos process easier I have interest now.

 A cropped version of the input image is here:

############################

https://github.com/andrewdefries/ComputerVision/blob/master/SampleImages/Test.jpg

  A picture can be found below showing an individual seedlings in a crop of
the large image. I did some thresholding and edge detecting in Fiji to get
some plant outlines, but I think some shape filters would be a good place
to start filtering objects for segmentation and local thresholding.

 I have tried some suggestions from your esteemed colleagues shown here:

###########################

https://github.com/andrewdefries/ComputerVision/tree/master/FijiMacro

  The question and answer below between Ignacio Arganda-Carreras and I
recapitulate the problem. Please read.

 ################ #####

Regarding your questions:

################ #####

 Q1) What do you exactly want to segment? Each individual plant or each set
of plants? I'm asking because I see that sometimes you have some of them
crossing each other.

 *He is referring to

###########

https://github.com/andrewdefries/ComputerVision/blob/master/SampleImages/TestLargeD_b4Particles.jpg

 Answer: The goal was to have seedlings grow separately, but as you noticed
some have crossed each other, or are twisted together. I would like to find
a function or if-else statement to separate these seedlings from each other
for further analysis.

  Q2) What do you want to measure? Number of plants? branches? lengths?

 Answer: Please see the image linked below this sentence with drawing on it
to follow my explanations. My goal is to extract data from 200-300 plants
per plate. I will now elucidate the colored lines drawn on said photo,
color by color.

 ##############

https://github.com/andrewdefries/ComputerVision/blob/master/SampleImages/SegmentationGoalsNColor.jpg

  Red lines

########

The red vertical lines represent the stem of the plant. I have been using
the length of this ROI using the hand polyline tool as my experimental
variable. I would like to find an automated way to decide where to put a
polyline vector in the center of the stem organ. Secondly, I would like to
use the outline of this organ to do a volume estimation, if possible.

 Purple lines

########

The purple lines represent the root, and I treat these similarly to the
stem organ. I ignore the arborization of the root now as it is not a
primary concern for my tracking at the moment. I use the length of the root
as my experimental value.

 Green oval

#######

This represents the cotyledon organ which is where the plant grows its new
organs. The angle of the geen oval with respect to the stem changes over
time, and I have been ignoring this angle change for lack of a proper
method to extract the data from the images.

 You can see this angle change significantly during the seven days of
observation. I put the symbols alpha and beta on the seedling on the right
to describe the polyline vectors I have been collecting. Seen in this
image:

##########

https://github.com/andrewdefries/ComputerVision/blob/master/SampleImages/SoilEmergenceA.png

 Notice on the second link below that the cotyledon opens up and peels
back. I would like to have a time series of the angles theta and gamma
shown. Also, it is desired to approximate the angles of change for the
cotyledon with respect to the other organs, if possible. Shown here:

##########

https://github.com/andrewdefries/ComputerVision/blob/master/SampleImages/SoilEmergenceB.png

  Yellow line

#######

 This line is showing the stem-root border. If you zoom or look at the
original photo you can see a tuft of root hairs, or sometimes a seed is
still attached at this point which can be used to anchor the measurements
perhaps.

 Cyan seedling

###########

 I hand drew this to show some more angles, areas that can be calculated to
inform my model. For instance, the royal blue area bound below the
cotyledon extending to the stem-root border should be represented in an
area under the curve graph. The same is true of the opposite side, shown in
dull green.

 We could make a bounding triangular ROI using the points 1,2, and 3 shown,
such as the connection of the cotyledon to the stem and the root-stem
border. Please let me know how to plan this for the workflow.

 Q3) Are these your final images or you're experimenting with the setup?
I'm asking because maybe you can facilitate the segmentation before imaging
by creating a more unified background (of one specific solid color) or by
using a grid paper with numbered squares, for example.

 Answer: I have a set ~500 of images I did without a grid, in the format
you see. I used a uniform LED transilluminator, but the plastic itself
creates excessive brightness near the edges. I have been cropping this out.

 Once I'm done extracting data from this image set I can re-build my
imaging apparatus. I'm also thinking of using high res flatbed scranners to
avoid artifacts of parallax that come with photography. I'm looking to
build a new platform to optimize conditions for the image analysis, but
have to get the data out to justify new projects.

 Thank you so much for your assistance. I look forward to your responses.
Please feel free to collaborate on github to facilitate executable code.

#####

https://github.com/andrewdefries/ComputerVision

#######

 Andrew Defries PhD



------------------------------------------------
Andrew Defries
Lab of Michael Pirrung
University of California Riverside
Botany and Plant Sciences
2150 Batchelor Hall
Riverside, CA 92521

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html