Help for mask / segmentation

[lechristophe]

Dear ImageJ specialists,

I'm relatively new to automatic image analysis. I have a bunch of images
of hippocampal neurons in culture with labeling of the Axon Initial
Segments (AIS, the beginning of axons). These AIS are bright, elongated
structures(typically 150-300 pixels with 10-20 pixels wide) and I would
like to automatically detect them in order to create a mask for further
quantification.

If I just threshold the images, I cannot isolate the full AIS from
images because I'd also get blob structures from the soma of neurons. So
I'm trying to find a way to "clean" the tresholded image : removing all
little particles and the big, round particles in order to only keep the
areas corresponding to AIS.

Unfortunately, I could'nt discriminate the AIS using the "Analyze
Particles" tool, I think because it's meant to measure round stuff
mainly. Even tweaking the "Circularity" did not help.

For those who would like to try ideas, I've uploaded exemples of raw
images (raw.tiff), thresholded images (mask.tiff) and manually edited
tresholded images (modified_mask.tiff). The manually edited images gives
you an idea of what I (ideally) would like to get as an output.


the adress is (2.6 Mo zip file containing 3 stacks of 4 images as TIFF
files):
http://www.christofunk.info/stuff/AIS_segmentation.zip

If you just want to visualize what I'm talking about, an overview image
is available at :
http://www.christofunk.info/stuff/Overview.jpg
The columns are : raw, mask, manually modified mask
The rows are : four different exemple images


I was also wondering if tweaking the image before thresholding (filters
and/or gamma to remove dusty labeling and backround) could help, or If I
could get the same results by using binary operation (erode, dilate,
close, open ?) after thresholding. I know very little about those
operation, particularily which to use for what purpose.

Thank you for your time and consideration,





--
Christophe Leterrier
Postdoc

INSERM UMR641 Neurobiology of ionic channels (Michael Seagar)
Ionic channels and neuronal polarity team (BŽnŽdicte Dargent)
IFR Jean Roche - Mediterranee University
51, bd Pierre Dramard 13916 Marseille cedex 20 France
http://ifrjr.nord.univ-mrs.fr/-Institut-Jean-Roche-?lang=en

tel: [phone]
fax: [phone]
email: [hidden email]

[Chu, Calvin-2]

Try applying a variance mask or mean filter (at least 9x9 pixels) to smooth out your image before thresholding.

Generally a variance mask works a lot better.