Need help understanding "Select Populations" Methods

The Customer

• October 07, 2020 14:57

Hello!

I have a quick question about Columbus' "Select Populations -- Linear Classifier" method. I am trying to quantify co-localizations of puncta [IHC of cortex] and also compare those co-localizations to puncta that only carry one type of marker. We are thinking of applying the following workflow before training the linear classifier:

1) "Calculate Image" method to add two different channels [Channel A and Channel B] together to get a composite image we'll call CompImage.

2) "Find Spots" of CompImage.

3) "Calculate Morphology Properties" of CompImage.

4) "Calculate Intensity Properties" of Channel A and then Channel B.

5) "Select Population -- Linear Classifer" of CompImage, where we ask it to identify puncta that shows up only in ChannelA, only in ChannelB, and then overlapping puncta. We train by toggling between the Channel views [right-side menu] and train visually.

My question is: If we train the linear classifer on CompImage, would that be able to differentiate between the different channels that went into creating the CompImage? i.e. If I denote that the population that I am training the Linear Classifier on is CompImage, but I only select puncta that show up in Channel A or Channel B [by using the right-side menu], will the Linear Classifer know that I've done that?

Or maybe a better question is: if I add channels together, does the Linear Classifier know to differentiate between the two different channels?

Thank you very much,

~Justine

0

• 

  The Customer

  • October 12, 2020 07:36

  Hi Justine,

  I ran your query past our applications team. Their response is below:

  ~~~~~~~

  The linear classifier only uses existing calculated parameters like intensity, morphology or texture.

  So if they want to separate Channel 1 from 2 they need to calculate parameters in these separate channels.

  In the CompChannel only greyscale information is available, so the classifier doesn’t know if there are different channels behind.

  So again, if the calculate first separate parameters for each channel and afterwards train the classifier as described it should work, but the CompImage/channel won’t help here, it can only be used to define the spot population.

  ~~~~~~~

  I hope this helps.

  Best,

  Nick

  0
• 

  The Customer

  • October 12, 2020 14:49

  Thank you very much! This does help.

  Best,

  ~Justine

  0

Please sign in to leave a comment.