Question about MICCAI dataset

[ghofranGh2022]

Hello @marianocabezas why did you use this file? What's its use?

[marianocabezas]

Hi @ghofranGh2022, which file are you referring to exactly?

[ghofranGh2022]

@marianocabezas train_challenge.py and also train_test_challenge.py ?

[marianocabezas]

train_callenge.py was meant to prepare the final model for testing (so it trained with all the available cases). train_test_challenge.py was meant for cross-validation, so it trained and tested at the same time. We tried a lot of different things for the challenge and cross-validation helped choosing the best option. Then, since we had to submit a Docker, we needed to train a single model with as much data as possible.

[ghofranGh2022]

Thank you @marianocabezas for this explanation, can i use this file in another code like nicMSlesions? so that dedicated approach can read data from MICCAI MS database?

[marianocabezas]

You could, but depending on what you want to do, you will have to change most of the code anyways. In the case of nicMSlesions, the code is a modified extension of that repository (Sergi and I worked closely together and then he adapted the code while I moved to other projects), so you might probably be better checking the code version of nicMSlesions directly...

[ghofranGh2022]

OK, thank you very much @marianocabezas , I used the data from the CHB-train database (Flair, T1,T2,lesion) to do the test using the baseline 2ch pre-trained model of the nicMSlesion approach but no results !!! I don't know what to do!! Similarly, when I did the training with the same data the same error is displayed! i use the Google colab.

[marianocabezas]

I have been following your issues with nicMSlesions (I also receiving them as part of the contributors). I think the problem might be with how storage works with Google colab. We always used it on our local machines or servers. Have you followed all of Sergi's suggestions? Like making sure the config files point to the right colab folders? From what I saw, it looked like a folders issue.

[ghofranGh2022]

ok @marianocabezas , thank you very much, I will check again the folder that contains the data as you told me.

[ghofranGh2022]

Hi @marianocabezas is this "https://github.com/NIC-VICOROB/nic_tools" another code that can be executed????? what is the use of this code?

[marianocabezas]

It's a bit more complicated than that. To summarise, the miccai repository was our first work with CNNs, then Sergi took that initial code and modified it to create the nicMSlesions repository. We realized that having code files and configurations of virtual environments was cumbersome for hospitals, so then we created a docker version which is what you can find in the nic_tools repository. There we also combined an old work of mine on lesion activity as a Docker too. The files there pull the Docker containers and run them on data. If you don't need much retraining they can be useful.

[ghofranGh2022]

it's understood, good job thank you @marianocabezas so, i am trying to apply nicMSlesions code, it is simple as that, and i hope it runs with me !!

[marianocabezas]

If you only want to test and do inference or fine tune to a set of images, nic_tools might be easier to use then. If you need to train from scratch or modify the code, nic_tools won't let you easily (as everything will be inside the container). It has a small GUI, but I think nicMSlesions has it too.

Not sure if that helps.

[ghofranGh2022]

ok @marianocabezas i am trying to do the test using nic-tool. thank you very much Mr @marianocabezas can i use nictool with google colab, because i don't have a high power machine??

[ghofranGh2022]

hi @marianocabezas can i run nictools under colab!!!!!!

[marianocabezas]

From my experience with colab, I don't think it can run a Docker container, but I could be wrong... The base code that uses Docker in nic_tools uses python, so you could give it a try.

[ghofranGh2022]

Thank you @marianocabezas @marianocabezas i have another question, know i am working with a gpu machine, i am running nicMSlesions, i am installing python 2.7.13, but when i go to install tensorflow it shows me a problem I don't know what I should do!!!

[marianocabezas]

Can you check what is the version of python that runs your pip? If you are using python3 it might be impossible to install older versions. Concurrently, python 2 support was dropped last year (if I am not mistaken) I am not sure if that also means that pip repos for python 2 were discontinued...

BTW, It might be better to open an issue on the nicMSlesions repo for stuff on it.

[ghofranGh2022]

OK it is understood I will do that Thank you @marianocabezas

[ghofranGh2022]

Hello @marianocabezas I need your help please, when i execute nicMSlesions with GPU machine , i find this error!!!! i don't know what i can do!! Thanks

[ghofranGh2022]

@marianocabezas and when i solve this by installing tensorflow and intel-tensorflow, i find this error

[ghofranGh2022]

And when i do "uninstall intel-tensorflow" i find this error, thanks @marianocabezas

[marianocabezas]

Having a look at the first two errors and the python traceback... It looks like you are using a python 3 distribution, while the code was developed for python 2 (and never ported to 3). There are quite a few big changes between python 2 and 3 that would probably break the code in some places (like the way zip works).

Unless you want to port the broken code to 3, and I would suggest using an issue on the nicMSLesions repository for that, the best solution would be to try to find a python 2 distribution with Tensorflow. But, since you have a GPU machine, I would rather suggest to use Docker and the Docker version in nic_tools. It should save you a lot of trouble with distributions and packages.

[ghofranGh2022]

Thank you @marianocabezas OK i will work with Docker

[ghofranGh2022]

@marianocabezas i need your help please at the same time, I ran the nicMSlesions approach on colab but the test execution does not stop and the folder created to contain the OUTPUT segmentation images is empty, I don't know why!!! Thanks.