Why doesn't Gabor Measure pick out z-line structure in cardiac myocyte?

We are trying to separate myocyte nuclei from fibroblasts using a stain for alpha-actin. Using a Lower Quartile Intensity measurement for the alpha-actin channel we achieve some selectivity for myocyte nuclei as shown in the attached image:

We would like to use a texture measurement as an additional way to select for myocytes by picking out the banded z-line structure. However, the Gabor measure does not seem to be correlated with z-line structure as indicated in the following image.

I used a scale of 3 pixels and a Gabor threshold = mean + std = 3.05

What options and values should we try in order to use the texture module to pick out the z-line structure?

Thank you for your assistance

Gabor-test.cppipe (29.1 KB)

Original Image:

(For some reason I was unable to upload original image. It may be too large to post.)

Hello! Have you tried using the new CPA features to extract which measurements could be most useful? You could attempt to classify myocyte vs non-myocyte.

I would expand the measurements on texture (more pixel scales, angles, etc) and then use the CPA classifier tool. This allows to a) classify them and (and simply use this classification) or b) you can display the top features and go back to optimising the pipeline correspondingly.

have fun!

Thank you very much for your reply.

I am concerned that the output from the Gabor filter does not make sense. The outlines selected appear random.