Giters

vessap / vessap

Experimental code and examples for: Automated analysis of whole brain vasculature using machine learning

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Looking for more documentation

JulianPitney opened this issue · comments

commented

Hey guys,

I read the preprint and readme and the pipeline looks really promising. We have some nice lightsheet whole brain vasculature scans and would love to use this on them.

Problem: I'm trying to figure out how to use the tool but can't find very much documentation on how to get going. Is there a plan to update the repo with more comprehensive docs? I also tried looking for the Code Ocean compute capsule but the link provided in the readme is "xxxxlink" which I assume is a typo. Let me know if there's somewhere I can go for more info on setting things up.

commented

Could you tell me whether I did it right and how I can understand the results? I ran python train.py with the defaults producing model1.dat. (This file seems to be different from your fcn_model.dat. Is it an operating system issue, which one did you work on?)
Then I ran python test.py data/1.nii.gz --model models/model1.dat producing 1_bins.nii.gz and 1_probs.nii.gz. What do these files show exactly and is it correct to use the bins file in the next step?
Afterwards I started python feats.py 1_bins.nii.gz and got the files 1_bins_bifs, cens & rads.nii.gz. How do I get from these images to the actual statistics of centerlines, bifurcation points and radius, is this also part of your pipeline or didn't you publish this part of the code?

btw: What is the purpose of the matlab scripts? They are not mentioned in the Readme file.

commented

Wondering if my docs question has been forgotten. Anyone there?

commented

Anyone there?

Doesn't look like that. :-(

commented

@JulianPitney, sorry for the so long ago question which has not been answered but I guess this might help someone else if not you. We thought the documentation we provided was enough but it seems that's not the case. We have to revisit this again. However, if you follow the steps used by @saskra above you should be fine.

commented

@jocpae please can you address the issue with the link to the capsule? Is it still available?

commented

@saskra The statistics on the features is what I think the MATLAB codes are for.

commented

@JulianPitney the code ocean capsule can be found here: https://codeocean.com/capsule/0511563/tree/v1 we also updated the link in the Readme.

commented

@saskra these answers consider the codeocean capsule (https://codeocean.com/capsule/0511563/tree/v1)

  • the train.py trains an exemplary model based on the image and label provided in the code ocean capsule, it is not the model we are providing as the fcn_model.dat has been trained on our large set of image label pairs (which are available on our website http://discotechnologies.org/). On code ocean we cannot host these large sets of data, neither can models be trained on such large data there.

  • when running the script feats.py you will get 3d volumetric data with the features (bifurcation points, centerlines, ..) as voxel information. Trivial numbers could also be returned, however for a detailed statistical analysis considering the brain regions, producing plots etc. you would have to use and adapt the Matlab scripts.