Retinal vessel analysis with Optical coherent tomography angiography

fiji
imagej

#1

I am interested in making a better way to contrast the retinal capillary structure around the fovea as Optical coherent tomography angiography shows it.
Actually I am trying to improve the capillary structure and eliminate the noisy background before to begin the process of vascular analysis.
So, I open this thread asking the best way to improve the image. Every way I tried erase some of the capillaries with a weak signal so I continue looking for the best.
I tried “Image…Enhance Contrast” with 25% of pixel saturation without Normalize or Equalize histogram.


#2

@valbayeros

It’s difficult for us to say what is best to do without seeing an image. Would you be able to either:

  1. attach an original image file here directly on this forum thread?

or

  1. share an original image file via a link (using something like Dropbox, etc.)?

This way - we can have a better idea of your image quality, etc. And if you can describe in a bit more detail the workflow you wish to carry out (subsequent measurements, etc) - perhaps we can advise you there as well?

eta


#3

[Normal OCTA 3000x3000 µm 464x464 pix]Normal Macula 464x464 pix 3000x3000 µm

This is an example of a normal Optica Coherent Tomography angiography.
Thank you very much for replying to my post. The workflow I do is the following:

  • I prepare the image: I convert to 8 bit and apply Threshold with several parameters. Some allow to choose with more specificity the capillaries and others choose with greater specificity the avascular zones.
  • I calculate the percentage of vascular area dividing the white pixels between the total of pixels per area and I report them by ROI based on the template ETDRS of the macula.

Another way to prepare the image that has been more specific is using the WEKA segmentation Trainable and I am more satisfied with its performance.
I have functional macros for both approaches. Although the WEKA approach seems to me more accurate, I can not avoid that when calling the classification model from the macro, it does not analyze the loaded image but an image that has been used to train the classification.