Image Segmenting overlapping variable regions (watershed with edge detection issues)

I want to segment multiple cells out of one image. I have images with the nucleus as markers and a separate cytoplasm image. There are multiple of these in one image. I have so far tried using the watershed method with markers. The nucleus image is the markers mask I have been using. However, this method has been inaccurate, due the stretches of overlay between different objects. I was wondering if there a specific way I can employ the watershed method to account for these type of images or if I should employ a sort of edge detection.

I was also thinking of perhaps going through each boundary and finding the points connected to each boundary when overlaying the markers on top of the original image. I would then try to employ a decider function for points connected to more than one boundary and see if I can decide which boundary those points belong to. It seems that the streaks coming out of the boundaries do follow a general path shape of the original boundary and are at parallel plane with the boundary.

Answer 1

[Image Analyst]

I think it's hard for even a human to decide where one object stops in that mess of filaments, and where the other begins. So I think you're best off having the user do it manually with imfreehand(). I know that's not what you wanted to hear but it could be your fastest route to a solution. You might spend hours manually tracing them vs. weeks or months trying to figure out how to do it automatically. If the image segmentation is your total project, then by all means try to figure it out. However if the biology is your interest and just need to get the cells identified by whatever means, and image analysis is just a means to the end, then just trace them manually and be done with it and get on with the biology.

[awezmm]

Thank you for your response. Image segmentation is the goal. Any suggestions on good ways to segment these type of images? Thank you.

[Image Analyst]

See my freehand masking demo, attached.

[awezmm]

I meant suggestions on a way to segment automatically, without having to select individual objects using imfreehand

[Image Analyst]

Well you didn't answer my question so I didn't know. I thought you maybe just wanted it segmented and that biology was your main purpose. But if the development of an image analysis algorithm is your main intent, and the biology is just the "use case" to help you do that, like your Ph.D. dissertation is on automatically segmenting cell images, then you'll have to develop something. Like I said it could take quite a while - more time than I can donate to you. Is this your graduate school thesis or dissertation, and you're going towards an imaging/engineering degree? (in which case you can't use some code I provide you) Or towards a biology degree?

[awezmm]

Thank you for your response. I will not be getting a degree from this and it is not a graduate school thesis or dissertation. It's just a side project. However, I do want to be able to segment automatically. Any suggestions for that?

[Image Analyst]

Try deep learning.