Comments (31)
Yes Slicer works well on the M1 Mac. I have also had success with NVIDIA's AI-assisted segmentation algorithm extension. However, your software appears to be much more straightforward interns of extracting volumes from.
from raidionics.
Okay so now I am seeing something. However tumor was only produced on the flair image and not the T1 Gad image.
from raidionics.
Okay, I see my mistake. I did not adjust the image modality - it was listed as T1-w not T1CE...Now I have the tumor annotations (2) under T1-CE.
from raidionics.
It would be nice to know a little bit more about your setup to debug this further.
Which hardware are you testing the software on? Are you using Windows/Ubuntu/macOS? Which CPU are you using on which computer?
For instance in my case I am using a Razer Blade Base 15 laptop, Intel core i7-10750H CPU, and Windows 10 Pro.
You can find this information from the System Information
.
from raidionics.
Absolutely, very sorry for the limited info.
Apple M1 Max, 10 core, 64 GB RAM, 32 core GPU, Monterey
from raidionics.
Absolutely, very sorry for the limited info.
No worries! :)
Apple M1
Oh, thats unfortunate. There are almost no softwares or libraries that are compatible with the new M1 CPUs. This has to due with the M1 CPUs having a completely different architecture from what people normally use and have used for quite some time (e.g., Intel CPUs).
I don't see us supporting M1 CPUs in the near future. It is really not up to us. Basically none of the libraries we are using in Raidionics support M1, hence, we cannot either.
I would suggest you to use a different computer to test the software. I'm assuming you want to use it in a study, or similar? We may be able to assist you in that, if you don't have any other computer to run Raidionics on.
from raidionics.
That is quite unfortunate to hear! Thank you for your response, I figure this might be the case. I will see if I cannot get my hands on an intel Mac or Windows and see how that works for my use case. Yes that is exactly right, I appreciate your offer. Thank you.
As an aside, would the slicer extension work?
from raidionics.
I will see if I cannot get my hands on an intel Mac or Windows and see how that works for my use case.
Please, test Raidionics on a different computer. It should work for most computers. If you keep having problems and unable to find a suitable computer, we can try to find other ways for you to perform your study :)
As an aside, would the slicer extension work?
Have you even been able to install Slicer? Thats surprising to me. I would assume Slicer would have issues, but perhaps they have made some hacks. One big library that Slicer use (or I think it uses) is called VTK, which I believe we have tested with M1 CPUs before with no success.
One of the fundamental libraries we use for inference in both Raidionics and the Slicer plugin is TensorFlow. Tensorflow is used for running the pretrained segmentation models and does not support M1 CPUs (or they do, to some minor extent, through a child library called tensorflow-metal, but only a single version, v2.5, which we do not support). So no, that will not work either.
Almost any computer should work with the software. However, we have also found that specific CPUs have issues with it. You can read about that in the troubleshoot section of the wiki:
https://github.com/dbouget/Raidionics/wiki/Frequently-Asked-Questions-(FAQ)#1-the-software-crashes-when-clicking-on-run-segmentation-or-run-reporting
Perhaps I shoud add information regarding the M1 CPU and make it easier for people to find this troubleshoot section in the main README. I can do that tomorrow :)
from raidionics.
@Stitchn99 I have now updated the FAQ in the wiki to include the M1 CPU issue:
https://github.com/dbouget/Raidionics/wiki/Frequently-Asked-Questions-(FAQ)
By NVIDIA software, i guess you mean MONAI-Label? Great software! I have been using it myself in some projects :]
Raidionics is handcrafted for tumor segmentation and does not require any training or coding yourself. You can simply just run the model on your data - get the segmentations and relevant features stored in a report.
But since Slicer works on your M1 mac computer, we should see if we can solve this in the near future. I will keep you updated :)
Will keep this issue open until this is eventually solved.
from raidionics.
Okay sounds great, so you are certain the slicer extension will not work for the same reason?
from raidionics.
Okay sounds great, so you are certain the slicer extension will not work for the same reason?
Yes, but you could try. The first library that will have problems is TensorFlow. We use TF v1.13.1, whereas tensorflow-metal uses tf v2.5, which are very different versions.
We have been considering updating, but we are unsure if it is just tensorflow that will have issues. I will assume ANTs also will not work. I made a comment on an issue about it in the GitHub repo:
ANTsX/ANTs#1235 (comment)
I also made a comment on the ANTsPy github repo (Python library, wrapped from ANTs):
ANTsX/ANTsPy#335 (comment)
Lets see what they say. If it works, we could push towards getting that working.
But for now, I would use a different computer for running Raidionics. I guess for your study this only needs to be done once on a set of images. Then you can just copy the exported output from Raidionics to the macbook after.
from raidionics.
More issues related to using M1 chip:
https://developer.apple.com/forums/thread/701358
from raidionics.
We will have more studies that could use the program. Do you think using a virtualization software such as parallels to run windows would work?
from raidionics.
We will have more studies that could use the program. Do you think using a virtualization software such as parallels to run windows would work?
Yes, that will work, as long as you have setup stuff on a different machine.
However, I would think it wouldn't be that challenging to get in touch with someone who has a compatible computer. Did you not find any? Even if you have two studies, you only need to do this twice.
Note that using remote desktop applications like Parallels still require you to have the data on the other computer, and not the M1. So I don't see why you cannot just do it on the other computer in the first place. It is quite fast to do.
BTW: New release today, hopefully!
from raidionics.
Okay so I was able to get access to another device. 12th generation i9, GTX 3070-Ti, 16GB DDR5 ram, Windows 11. Now I am getting the error that "your cpu supports instructions that this tensor flow binary was not compiled to use: AVX2"
from raidionics.
"your cpu supports instructions that this tensor flow binary was not compiled to use: AVX2"
AFAIK, this is a warning
and not an error
. What this means is that the program will still run but TensorFlow inference is running a little slower than it could do, if TensorFlow was compiled for AVX2. This is not a problem. You can still use the software. Have you tried?
We just made a new release. Please, try downloading the new release:
https://github.com/dbouget/Raidionics/releases/tag/v1.1.0
And then see the tutorial wiki for how to use it (get started):
https://github.com/dbouget/Raidionics/wiki/Tutorials
Click on the images to see the tutorial video. This should redirect you to YouTube.
Let me know if you still have issues.
from raidionics.
Thank you for the response. To confirm - do I need to install any dependencies and/or install tensor flow (this is a new computer)
from raidionics.
Thank you for the response. To confirm - do I need to install any dependencies and/or install tensor flow (this is a new computer)
No. The program should just work out of the box. But if not, please, let me know such that I can assist you.
from raidionics.
Unfortunately, after uploading the data and running segmentation per your instructions. The process runs but produces no result.
from raidionics.
The process runs but produces no result.
I believe the results may be generated, but they are not disabled by default. This was done as each model produce multiple different segmentations (tumor, brain, and different atlas-generated structures).
See this video for how to choose which result to show:
https://youtu.be/kU_Mt5tUGTA?t=24
from raidionics.
Its a new software, which you have never used. It is expected that some features might not be very clear to you, but great that you are telling us what is not intuitive.
We should try to find ways to improve the GUI to better guide the user :) So it is great that you are telling us about this!!
Let me know if you are having more problems.
from raidionics.
Okay, I see my mistake. I did not adjust the image modality - it was listed as T1-w not T1CE...Now I have the tumor annotations (2) under T1-CE.
Oh yeah, note that we do not automatically detect which sequence type is used. This cannot be done rigorously through the metadata. However, we have trained a model which can classify sequence types, and are planning on adding it to the next upcoming release. Stay tuned!
I would recommend you to star the repo, by clicking on the button of the top right in the repository:
That way you will get notifications on new releases (and other stuff), and you will have updates in your github feed. This is how it looks for me (but note that I follow lots of projects, so you can see that other non-Raidionics-related stuff is shown there).
from raidionics.
Thank you for this info. Unfortunately, I have run into another issue with the gui. When attempting to open an empty patient record, it appears to be throwing an error on the command prompt and failing to do so.
from raidionics.
When attempting to open an empty patient record
What do you mean by "empty patient record"? Like loading the result saved on disk, or do what?
from raidionics.
This was occurring during the first step of loading data. However, I was able to correct this (through some user-level issues) and it is working as it should now.
from raidionics.
This was occurring during the first step of loading data. However, I was able to correct this (through some user-level issues) and it is working as it should now.
Note sure if I get what the problem was. Could you take some snapshots of what you are trying to do?
Also, since this is not a macOS-related issue. Perhaps we should move this discussion to a new issue (such that it is more easier for others to find this Issue). Could you make a new one?
I will keep this open until M1-support is eventually added.
from raidionics.
@Stitchn99 But most importantly, if I understand correctly, you are able to run an analysis on an image and the result is generated now? Have you tried running an analysis on a batch?
from raidionics.
@Stitchn99 Just letting you know, that we will have a test version of Raidionics made available for you to try, that works on M1 chips. Do you still have such a Mac to test on?
from raidionics.
We have made a pre-release ready for you to try:
https://github.com/dbouget/Raidionics/releases/tag/v1.2.0-alpha
We have run tests on various machines and setups and it seems to be working well. I am therefore closing this issue. If you are experiencing any other issues, please, let us know and we can try to help you further.
from raidionics.
Related Issues (20)
- Update hashtags HOT 2
- Improve wheel sprites
- Minor UX improvement HOT 1
- .app icon not showing on MacOSX HOT 2
- Start Menu icon not showing on Windows HOT 2
- Wrong style sheet on macOS HOT 2
- Memory leak in batch processing HOT 4
- Suboptimal behaviour of circular progress widget
- Bug: Unable to kill process when running HOT 2
- Circular progressbar not reset when starting new segmentation HOT 1
- Enhance about
- Processing on macOS hangs HOT 2
- PyInstaller packages ALL local python versions when building HOT 5
- Feature: 3D renderer support HOT 1
- segmentation on Mac M1 vs macOS Catalina HOT 5
- FLAIR segmentation not running for low-grade glioma mode HOT 9
- Bug: Automatic models folder deletion after re-installation HOT 2
- DICOM RTstructure segmentation export
- Feature: Interactive prediction thresholding
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from raidionics.