Circles app

Wondering if anybody can help me with the circles app. I am using the app to help count blood cells on a blood smear. I always seem to end up with some for which circles overlap too much. If I increase the minimum c-c distance I start losing identification of cells (with circles of roughly equivalent size) in close proximity before I get rid of smaller circles (that are not actual cells anyway) that are identified mostly within larger circles of true cells. I can filter most out be size during the count but would like to know if there is a better way to do it.  Thanks.

Answers

  • Circles app has limitations and detects only perfect circles, though there are many parameters to play with. Can you attach your image, so I can check if the app can be used with it?

    Thanks,

    Yuri

  • Yuri, the circles don't need to catch the edges perfectly and actually it is mostly helpful if it doesn't. This is taken from an image of a blood smear of mouse with malaria. I want to count parasites inside the red blood cells only and when I do a second count using smart segmentation and then look at data with parent-child view it nicely does not count the cells where the parasite is stuck on the outside of the cell because it is not fully within the circle. Jeff has worked with me a lot on using two stage segmentation and that works fantastic as long as infection burden is not too high.  But as the number of organisms in the blood goes up there are more free-floating parasites in the blood that get stuck on to the cells and error in counting those cells goes up. I also don't care how many parasites are in an individual cell....just whether there are any so if, for example, a red blood cell has three parasites within it and only one is counted because the other two touch the circle of the parent then it is still counted. White blood cells also increase but I can filter those out with counting criteria most of the time.  There is also a bluish tinge to the smears with the higher numbers of parasites due to high levels of gamma globulins in blood that pick up the stain. This screws heavily with getting good segmentation as it makes cells that are in close proximity harder to split when counting.  I am checking into some specimen preparation alterations that might reduce this because the method that Jeff help me with is superior and easier to alter to adjust for slide variations...but I thought it was worth comparing the two.


  • The image looks good , many cells are detected. Can you post the original image, without measurements, so it can be tested?

    Thanks,

    Yuri
  • 2019-12-02-155150

    All --

    I am adding material to this discussion to hopefully assist anyone else looking at the FIND CIRCLES TOOL within IMAGE-PRO.

    I have an image which contains a collection of circular features.  A cropped section of that image is below (JPG) and attached (TIF).



    I would like to measure the well defined circular features something like what is shown below.


    I have attempted to use this cropped section of the image to set the parameters for the FIND CIRCLES TOOL as shown below.



    I have not been able to find a configuration for the FIND CIRCLES TOOL which does a good job of locating the circles in this image.  The configuration shown above and below is about the best that I was able to achieve.



    Is there a configuration for the FIND CIRCLES TOOL that will locate all or most of the circular features that I identified manually with the BEST FIT CIRCLE TOOL?

    Thanks.

    -- Matt
  • edited December 3
    Hi Matt,

    We have a specialized Circles app that uses circular Hough transform and more options to detect circles properly. You can download that app from our library: https://www.mediacy.com/support/imagepro/appcenter/circles-application-detail 

    Here is how the result looks if I resize your image to 25% and use the attached options (same as on the screen).


    Yuri
  • edited December 5
    2019-12-03-093737

    Yuri --

    Thank you for the guidance on this.

    I will weave the information you have provided into the PROOF OF CONCEPT that is being done for a potential IMAGE-PRO CUSTOMER.

    Thanks.

    -- Matt

    2019-12-05-110609

    Yuri --

    It looks like my POST COMMENT on the message above did not work completely because the COMMENT appeared here when I entered this discussion.  Sorry about that.

    I was able to duplicate your results with the 25% SCALED IMAGE.  Please see the SCREEN CAPTURE below from my computer.

    I will work with this and see if I can weave this into the PROOF OF CONCEPT that we are working on.

    Thanks again.

    -- Matt



Sign In or Register to comment.