search

uncbiag / uniGradICON

The official website for uniGradICON: A Foundation Model for Medical Image Registration
Apache License 2.0
82 stars 4 forks source link

WSI Registration Support #23

Open eshasadia opened 1 month ago

eshasadia commented 1 month ago

Hi, Is there anyway to try this model on whole slide images? I am currently working on restrained and consecutive wsi slides and would like to see the result of this algorithm

lintian-a commented 1 month ago

Hi @eshasadia ,

Thanks for your interest on trying uniGradICON on your data!

I haven't worked with WSI slides before and haven't tested uniGradICON on WSI slides. Thus, I am not sure how it would perform. There are currently three ways to run uniGradICON/multiGradICON:

What file format are the images stored in? And are they 3D images?

If none of the above interfaces work in your case, what customizations are needed for the WSI slides?

marcniethammer commented 1 month ago

Also, uniGradICON is trained for 3D whereas I assume your WSI are in 2D. GradICON can be trained for 2D as well (but we don't have a general model for it).

eshasadia commented 1 month ago

yes, I have multi stained 2D WSI files, is there anyway to try this model on them.

marcniethammer commented 1 month ago

This model is currently focused on 3D. For 2D a starting point could be #17 But ultimately it would be useful to train a dedicated 2D model.

HastingsGreer commented 4 weeks ago

GradICON can train on images up to 4096 x 4086 on an A100. Here is a notebook demonstrating this on a toy dataset:

https://colab.research.google.com/drive/1BoWHAYhMwzSfj5BKxcEOSbcYUjm9ehof?usp=sharing

I would be very curious to know how well this training works if you swap in your data for the toy dataset used in the notebook, and happy to support if you run in to issues (or success!)