Is it possible differentiate objects according to shape?

cellprofiler

#1

Hello, I have just started to use the CellProfiler.
In the image there are the pollen grains in different polar views. We need to separate the pollen grains (shape can be good trait) into 3 groups how they are marked in the image. Then we need separately measure the grains from these groups. Is it possible to use the CellProfiler, and what kind of module I can use for it? Thank you for your advice, janka.


#2

Hi Janka,
I guess you can try to use PrimaryObject detection and Measure as many Features as you can (select them all).
Then you can use Cell Profiler Analyst to categorize your pollen grains.

One very important step, is to categorize in a robust way.
On your 1st/top image, it’s a bit confusing for me because the bottom blue pollen grain looks more like the one in the green circle ( with 3 poles) rather than the top/other blue circle.

“If you give confusing information to the classifier, you will get confusing result.”

Best,

Romain


#3

Hi Romain, thank you very much for your fast reply. I have just started to work with Cell Profiler so your respond help me to know that I can continue. Do you know should I have to prepare the images first of all through the Cell Profiler and after that can I use the Cell Profiler Analyst? The aim is to categorize the cells according to shape and then measure these cells in separated groups. Thank you also for your notice about the confusing information for classifier. Yes we will set the information as precise as possible. Thank you very much once more. All the best janka


#4

Hi Janka,

Do you know should I have to prepare the images first of all through the Cell Profiler and after that can I use the Cell Profiler Analyst?
it’s exactly how it works ! First, you make a CellProfiler pipeline. Isuggest to use “detect primary object” to find your particles, then you measure some features (intensity, texteture, etc) and you export these measurements. I recommend you to use the SQLite option and Select the option “Create a CellProfiler Analyst properties file?”.
In a second step, you start the Analyst, open the properties file, and you start classify your particles in different categories.

As an alternative, you could also consider using ilastik
You’ll need to create a project with pixel AND object classification.
First, you draw some pixels defining background VS particles.
Please find below some advices from a previous post :

In a second step, you’ll have to classify some objects in categories.

Best,

Romain


#5

Hi Romain, your last notes where very helpful and encouraged me a lot. I have worked on the topic (classify the pollen grains – polar, equatorial view) and I have created SQLite database through the Cell Profiler pipeline. I have measurements for - size, shape, intensity and texture of identified objects. Also I have already started with classifier in CPA. And here comes up my further questions.

  1. When I am setting up my training set and then look for evaluation results - through the Confusion matrix or Cross-validation accuracy - What percentage (results) is enough to start score all?
  2. When I have the results – is it possible to display plots for different measured parameters but selectively for objects from positive or negative bins?
    Thank you very much in advance for your answer and help. All the best janka

#6

Hi Janka,

The closer to 1 the better :wink:

I’ve to say that I didn’t use the CPA plots functions that much, maybe the @CellProfilerTeam could answer ?

Best,

Romain


#7

Hi, I’m totally a new learner, I also want to do differentiating jobs so as to classify two cells using machine learning methods attached to CPA. I have three image sets, two types of pure cells and a mixed one. I want to train machine with two pure cells and score the mixed one. How can I use CP to analyze those image sets in only one pipleline. Now I have three pipelines analyzing three cells, but I find it difficult to combine them into a one pipeline cuz I can not start CPA with three property files. Really looking forward to your reply, thank you! It is very urgent to solve this problem, thanks again.