Image analysis

[SaharFarhangi]

I wanted to analyze the g-ratio but I faced this issue: FileNotFoundedError. I tried to upload and check the g-ratio of my friend's image, there was no issue.

[Stoyan-I-A]

Hello @SaharFarhangi,
In order to compute the morphometrics (such as the g-ratio), the software requires the segmentation masks of the axons and myelin of your image. Do you have these segmentation masks? If so, are they in the same place as your microscopy image and what are their files called?

[SaharFarhangi]

My problem is that it cannot recognize my image.

[Stoyan-I-A]

@SaharFarhangi From the screenshot that you showed, it seems that the software is looking for a file called 155A4_seg-axon.png. The suffix _seg-axon.png is the one that AxonDeepSeg uses to save and identify the axon segmentation mask.
If no segmentation mask is present in the folder with the microscopy image, the morphometrics cannot be computed.

Could you describe the previous steps you performed before getting this error? Maybe this could help figuring out what went wrong.

[SaharFarhangi]

Sure.
Turn on ADS plugin
Drag and drop grayscale inverted image into napari
Select model > rat TEM model
Settings > set zoom factor to 1 or 2 (I checked different zoom factor)
Apply ADS model > set pixel size to 0.069 micrometers per pixel

When I drag my colleague's image, it works without any errors.

[Stoyan-I-A]

@SaharFarhangi Thank you for the response. I have a few ideas on what could be the problem. I have a few questions:

1- What are the dimensions (in pixels) if your image?
The TEM model that we provide was trained with resolution of 0.01 micrometers per pixel. AxonDeepSeg would resample your image (which has a resolution of 0.069 micrometers per pixel)) before applying the model to try to have a resolution similar to what it was trained on, but in your case this might make the image too small and the software will fail to apply the model because the resampled image cannot fit a patch.

You could try to segment with a zoom factor of 0.14 and see if it helps.

2- If you get the error again, can you copy and paste the whole content of the terminal? This is useful information for the team to better understand where the error happens and if there are any indication of what could be the cause of the error.

[SaharFarhangi]

This is the dimension.

[Stoyan-I-A]

@SaharFarhangi Thank you for the information. As mentioned, you could try to segment using a zoom factor of 0.14

If you still get an error when trying to segment, please copy and paste the whole output of the terminal.

[mathieuboudreau]

Another option to debug would be to send us a copy of the image, and we could likely quickly find the optimal settings. The histology images look quite great, so I don't expect any issues getting the segmentation to work well and subsequently the morphometric values like g-ratio

[SaharFarhangi]

Thank you for your assistance. Zoom factor 0.14 is working.
I have attached my image. Some axons are not fully filled with blue color (e.g., numbers 65 and 105). Should I fill them in? Additionally, I encountered some delamination, causing ADS to mistakenly identify them as separate axons (e.g., number 74). What do you suggest I do?

[mathieuboudreau]

Some axons are not fully filled with blue color (e.g., numbers 65 and 105). Should I fill them in?

You can either to manual corrections (removing the myelin that was segmented inside the axon, and then filling it all in) or omit that axon in your analysis of the morphometrics (you can do so by looking at their ID numbers in the image above, and then not including those rows in the morphometrics file).

Additionally, I encountered some delamination, causing ADS to mistakenly identify them as separate axons (e.g., number 74). What do you suggest I do?

Same thing, manual correction or omit; the choice is up to you, if you only have a few images to do then manual corrections is probably the best route, if you have many images then filtering them to be ignored in the morphometrics file is likely more time efficient

[SaharFarhangi]

Hi, I have encountered a recurring issue with my images, particularly with the larger axons, such as axon number 65. In this specific axon, there is a hole in the myelin and the myelin is not fully closed around the axon. I'm concerned about how to proceed without compromising my data.

If I delete this axon, I risk losing all the larger axons, which are crucial for my analysis. However, if I attempt to correct it by removing the segmented myelin inside the axon and filling it as an axon, I'm worried it could affect the myelin thickness measurements. Conversely, leaving it uncorrected might impact the g-ratio and other related metrics.

Could you please provide some guidance on how to handle this situation?

[SaharFarhangi]

Can AxonDeepSeg be used to visualize different myelin thicknesses with different colors, similar to the attached image?

[Stoyan-I-A]

Unfortunately, we do not have this feature in AxonDeepSeg at the moment.

[hermancollin]

well I think we have a PR for this (more or less) from @mathieuboudreau

[mathieuboudreau]

@hermancollin do you mean #781? I don't think my PR does that at the moment, but it could probably be integrated fairly easily

[hermancollin]

Yes that's what I meant. Weird, I thought we discussed "heatmaps" like in the original AxonSeg software but maybe we missed that.

[SaharFarhangi]

Given the 3 different magnifications I have for Axondeepseg analysis-4800x, 6800x, and 9300x- and considering that delamination is evident in my images, which magnification do you recommend based on your experience?

[hermancollin]

Hi @SaharFarhangi. If I understand correctly, the image you showed above was at 9300x, correct? If that's the case, it looks pretty good so you could continue with this magnification.

At the moment your images have a pixel size close to 0.1 um/px. In theory this is ideal for our TEM model.

[SaharFarhangi]

Hi, I have encountered a recurring issue with my images, particularly with the larger axons, such as axon number 65. In this specific axon, there is a hole in the myelin and the myelin is not fully closed around the axon. I'm concerned about how to proceed without compromising my data.

Based on @mathieuboudreau suggestion, If I delete this axon, I risk losing all the larger axons, which are crucial for my analysis. However, if I attempt to correct it by removing the segmented myelin inside the axon and filling it as an axon, I'm worried it could affect the myelin thickness measurements. Conversely, leaving it uncorrected might impact the g-ratio and other related metrics.
Does AxonDeepSeg detect myelin folds and correctly quantify the myelin sheets without including the area in the fold?

Could you please provide some guidance on how to handle this situation?

[mathieuboudreau]

Hi @SaharFarhangi,

Based on @mathieuboudreau suggestion, If I delete this axon, I risk losing all the larger axons,

No, you could selectively (by using the axon IDs) simply omit this one axon from your analysis

However, if I attempt to correct it by removing the segmented myelin inside the axon and filling it as an axon, I'm worried it could affect the myelin thickness measurements.

Hum, I don't think this is a worry - just erase the mask for that axon and manually segment it, it should then provide regular values for myelin thickness and size.

In general, your segmentation image looks quite good for most axons, this type of artifact can happen when there is some other dark features visible in the axon (i.e. if it doesn't look "regular" like what was in our training data).

The choice between using image as is but omitting IDs from the analysis or manually corrected the masks, to me it boils down to how many images do you have. If you only have a few (eg 10-30), then manually fixing these instances seems feasible time-wise. If however, you were to have hundreds of images, then I would first try omitting these from the statistics (reporting this in any publications), and you could select a few random images to see if there's a common feature in the ones you omitted (eg size, g-ratio) and do some aditional analysis to see if omitting these substantially impacted your analysis/results.

[SaharFarhangi]

Hi, thank you for your response.
I am analyzing the optic nerve, which include both small and large axons, across approximately 10 images for each sample. The issue with segmented myelin occurs specifically in the larger axons. I know I can delete the axons with the ID, but if I do this, I would have to delete all the large axons, which is a significant issue as they are crucial for my analysis.
Would you please explain more about it: just erase the mask for that axon and manually segment it, it should then provide regular values for myelin thickness and size.
Additionally, could you please take a look at axon number 58? It does not appear to be a separate axon but rather a segmented part related to axon number 59? How should I handle this situation?
For your review, I have attached 2 images at x6800 magnification, showing the segmentation before and after correction. I deleted the segmented myelin and filled it as an axon. Is it correct?

[hermancollin]

@SaharFarhangi hello!

Would you please explain more about it: just erase the mask for that axon and manually segment it, it should then provide regular values for myelin thickness and size.

If the model misses some axons, you can correct the mask before re-applying morphometrics. The manual correction you showed is a good example.

Additionally, could you please take a look at axon number 58? It does not appear to be a separate axon but rather a segmented part related to axon number 59? How should I handle this situation?

Yes, unfortunately small glitches like this happen sometimes. You can correct the myelin mask + remove the spurious axon 58 in the axon mask.

For your review, I have attached 2 images at x6800 magnification, showing the segmentation before and after correction. I deleted the segmented myelin and filled it as an axon. Is it correct?

Yes, overall this correction looks good. However I would be cautious when deleting the myelin to avoid under-segmentation.