Find Myelin Volume Fraction from axon diameter

[tanguyduval]

Use a nice SEM image. Find the relation gratio=f(diameter) (try to find refs in literature also). Then use this relation to find myelin thickness of each axon.

We will use this relation in case of bas myelin segmentation (e.g. myelin exploded/ myelin figures/ resolution too low)

[tanguyduval]

Please put a plot of gratio=f(diameter) in this discussion

[alzaia]

If I take my last results (with quadratic DA) its the same plot as the one we plotted last friday (see screen capture). Do you want me to take a cropped one & be sure I have no false positives & false negatives? Anyway, I have to do a manual one someday (no FP or FN) so I can also use it for my other validations.

[tanguyduval]

Discard problematic points (on the right). Use multiple images. Try to linearize: plot log(gratio)=f(log(diameter)).

2015-11-16 15:54 GMT-05:00 alzaia notifications@github.com:

[image: screen shot 2015-11-16 at 3 49 54 pm] https://cloud.githubusercontent.com/assets/14980394/11194396/c4ad483a-8c79-11e5-861d-1a0596ea577e.png If I take my last results (with quadratic DA) its the same plot as the one we plotted last friday (see screen capture). Do you want me to take a cropped one & be sure I have no false positives & false negatives? Anyway, I have to do a manual one someday (no FP or FN) so I can also use it for my other validations.

— Reply to this email directly or view it on GitHub https://github.com/neuropoly/axon_segmentation/issues/18#issuecomment-157167224 .

[alzaia]

Found this in the literature. So its supposed to look like this?

[tanguyduval]

exactly! Great! put the REF please

2015-11-17 10:39 GMT-05:00 alzaia notifications@github.com:

Found this in the literature. So its supposed to look like this? [image: screen shot 2015-11-17 at 10 31 28 am] https://cloud.githubusercontent.com/assets/14980394/11216111/8898641c-8d17-11e5-9124-6a58517da77c.png

— Reply to this email directly or view it on GitHub https://github.com/neuropoly/axon_segmentation/issues/18#issuecomment-157407265 .

[alzaia]

Ikeda 2012 -Relationship between nerve conduction.pdf

[alzaia]

When I keep all our data (combo of 4 different axon segmentations), we have a 'bump', probably due to biais from our segmentation (& maybe from the myelin thickness too). So keeping the whole dataset is not great, as you can see in the left pic. But after some data selection, we get the log regression appearing on the right pic.

[jcohenadad]

it's quite confusing to have axon diameter in pix-- i would convert to um

[alzaia]

yeah I know but in the axon segmentations I used as dataset, I dont have the pixel size. But I did a quick conversion (for the resolution I used in these images, 1 pix = 0,185 um) so I get this :

[alzaia]

But again, in the data I had we mostly had small axons, & the paper I was referring to is about sciatic nerves, so hard to compare.

[tanguyduval]

Something is wrong. If you used the following images I see lot of axons

5um (>4px by the way):

2015-11-17 17:42 GMT-05:00 alzaia notifications@github.com:

But again, in the data I had we mostly had small axons, & the paper I was referring to is about sciatic nerves, so hard to compare.

— Reply to this email directly or view it on GitHub https://github.com/neuropoly/axon_segmentation/issues/18#issuecomment-157533674 .

[jcohenadad]

we do have pixel resolution from SEM data-- the scale bar is always indicated. BTW: I agree with Tanguy something is wrong with your conversion-- visible axons should be in the 1-10um range

[alzaia]

This is the result with the whole dataset after updating the x axis (um), it goes from 0 to 10 um :

[tanguyduval]

Much bettter! Different color for different colors/contrast please!

2015-11-17 18:07 GMT-05:00 alzaia notifications@github.com:

This is the result with the whole dataset after updating the x axis (um), it goes from 0 to 10 um : [image: screen shot 2015-11-17 at 6 06 11 pm] https://cloud.githubusercontent.com/assets/14980394/11227763/1a1f5164-8d56-11e5-9a67-72f76f7aacc4.png

— Reply to this email directly or view it on GitHub https://github.com/neuropoly/axon_segmentation/issues/18#issuecomment-157539141 .

[tanguyduval]

We can notice the inconsistency for larger axons: their myelin is more exploded and present "figures"

2015-11-17 18:10 GMT-05:00 Tanguy DUVAL duvaltanguy@gmail.com:

Much bettter! Different color for different colors/contrast please!

2015-11-17 18:07 GMT-05:00 alzaia notifications@github.com:

This is the result with the whole dataset after updating the x axis (um), it goes from 0 to 10 um : [image: screen shot 2015-11-17 at 6 06 11 pm] https://cloud.githubusercontent.com/assets/14980394/11227763/1a1f5164-8d56-11e5-9a67-72f76f7aacc4.png

— Reply to this email directly or view it on GitHub https://github.com/neuropoly/axon_segmentation/issues/18#issuecomment-157539141 .

[jcohenadad]

when i said "1-10um" range, that was a guess estimate, not the actual distribution of axons. You should look at the exact scale bar.

p.s.: guys-- could you please not "reply" to emails, otherwise it accumulates in the github thread-- thanks

[alzaia]

Yes I used the exact scale bar. The bigger axons on the image are around 10 um (see screen capture below i.e. one of the biggest axons) :

In my case, I'm replying directly on Git. What should we do instead?

[jcohenadad]

nice! results make much more sense now. Discrepancies from the Ikeda paper probably comes from the fact that you have almost now visible axon with diameter less than 1um.

However I am wondering how Ikeda et al. could measure g-ratio with axons less than 100nm using optical imaging!

p.s. i was referring to Tanguy for the reply ;-)

[tanguyduval]

Very good point! Not specified in the paper. By the way I'm a bit suspicious on some of their data point.. They are strangely aligned no :-)

[alzaia]

I tested the curve fitting with different datasets I had for the SEM data. Its very similar for all datasets, our g-ratios dont vary too much (mostly stay between 0.4 & 0.7). Its easier to get the smaller axons when using a discriminant analysis, thats why the fit has a bigger range.

[jcohenadad]

Aldo, I don't understand: when you say "different dataset" do you mean different images? if not, what do you call different datasets?

[alzaia]

No I mean same data but different segmentation processing in fact, sorry :P But I have the same pattern for a 1500x SEM one (small dataset), but this one is manually corrected, so no false positives :

Just wanted to show that the processing doesn't matter that much in this case, all methods give very similar g-ratio VS axon diameter graphs.