Segmentation issue with my images: small green dots called mitochondria not accurately being detected in code

I’m having an ongoing issue with my code: it is not properly detecting the mitochondria. When I zoom in, the mitochondria is still not segmented enough (the current segmentation sometimes counts multiple mitochondria as 1). Then it affects the other measurements: circularity and average size. I have been wracking my head trying to figure out how to segment it better but I don’t know how. Would someone be able to show me How to perform a better segmentation? My image is attached. Thank you, please help.

my issue is not capability, its accuracy. The count is fine but some of the segmentation is not strong (i.e. some of the green blobs are turned into strings instead of closed area objects and the feret size is larger than expected although I converted everything to microns), would this be something you can look into? Thanks for considering!

Answer 1

[Image Analyst]

Perhaps try this:

https://www.mathworks.com/matlabcentral/fileexchange/126325-tree-detection-with-decremental-circle-fitting-algorithm?s_tid=ta_fx_results

[Chanille]

@Image Analyst Thank you, i was able to detect them with another functionbut thank you for your effort.

Answer 2

[Image Analyst]

Sometimes you can't measure what you think you'd like to measure, so you have to go with what you can measure. For example if you can't count the number of mitochondria you can compute the sum total of all the fluorescence. And maybe that will correlate with whatever you're hoping it would. For example, maybe the sum or mean gray level (which you can measure) correlates just as well with drug activity, drug concentration, or drug efficacy as the number of mitochondria (which you can't measure). If you can't segment out the individual mito, then I suggest you try this approach.

[Chanille]

Can you show me how I would do that?

[Image Analyst]

greenChannel = rgbImage(:, :, 2);
fluorescenceIndex = mean(greenChannel)

[Chanille]

@Image Analyst can i send you my code for you to review? i am happy to compensate.

[Chanille]

@Image Analyst my issue is not capability, its accuracy. The count is fine but some of the segmentation is not strong (i.e. some of the green blobs are turned into strings instead of closed area objects and the feret size is larger than expected although I converted everything to microns), would this be something you can look into? Thanks for considering!