Count "Circles" of an PES Fiber membran generator

KNIME Extensions
1

Hello all,

I was asked to do Image Processing with the picture of a generator filled with PES Fiber membranes. So till I have no clou about KNIME, I was a bit overwhelmed. Till now I used the given workflow for cell counting, but it stops and got a problem with the "Waehlby Cell Clump Splitter" with the notification you see in the file. It seems like I miss a second collumn with the information needed. Do you have an Idea what is the problem?

Thanks

2

Hi Deathlezz,

The error indicates that you are missing a column that contains images (ImgPlusValue is the internal name for images).

I have a few resources for you to get an easier start with KNIME:

  • The paper:  http://www.uni-konstanz.de/bioml/bioml2/publications/Papers2016/Dietz2016.pdf is a comprehensive tutorial on how to use KNIME for bio image analysis.
  • We have example and tutorial workflows available in the nodeguide section of the website, I think especially the High Content Screening workflow could provide you with some guidance

If you have any more questions feel free to ask them here!

best,

Gabriel

3

Hey Gabriel,

thank you so far. Now I got the problem that the "easy" picture I test the workflow with is segmented in to many parts. Ich just want to count and segment the "circles" at the end. Is there a way to reach the goal. Cause like it is now, the counted number is much higher than I want it to be.

Best regards!

4

There are two ways you can deal with this:

  1. You can use the Labeling Filter  node to filter out segments that touch the border of the image, this should get rid of the large rectangular regions.
  2. You can use the Image Segment Features node to calculate the circularity feature and use it to filter out only the round enough segments.

I have attached an example workflow that demonstrates this using the image you uploaded. I also included how I got to the selection criterion by using a decision tree.

best,

Gabriel

 

5

Thank you, so far the segmentation works, but at the end I count the "Number of Unique Labels" and it counts 24. I do not know how to make it count the values so that the result is correct (8)?

Best regards

6

You are probably running your analysis on a 3 channel image (RGB colors), resulting in each "circle" being detected in each channel which results in 3x8 = 24 unique labels. You have several options to deal with this:

  • Flatten your image to one dimension using the Projector node, using the projection operation AVG_INTENSITY and the projection direction Channel, worked very well for me.
  • Only work on one channel, you can split the image using the Spliter node and use the Column Filter to remove the dimensions you don't need (you can see how I did that in the Create Labeling metanode in the  example workflow I added to my last post.
7

Is there a possibility that KNIME can recognise a shape. In a way that it only holds shapes that are close to circles?

8

Yes you can, I already included this in my workflow, see my earlier answer:

You can use the Image Segment Features node to calculate the circularity feature and use it to filter out only the round enough segments.

The circularity feature is based on a mathematical function that tries to measure how "circle-like" a shape is. If it is not sufficient to distinguish between your shapes, you will need to find other features that do, take a look at the Geometric features available in the Image Segments Features node.

9

I segmented the Image and got the circularity, but how do I tell the software to just use the ones with a certain value? Sry if it is described in the Workflow you posted, I can't open it, because of IT "safety issues".

10

I just used a Row Filter node, see following image for the values I used:

705×643

11

I just used a Row Filter node, see following image for the values I used:

705×643

12

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.