search

CBICA / Deep-LIBRA

11 stars 4 forks source link

Sample images #3

Open catalinabustam opened 2 years ago

catalinabustam commented 2 years ago

Hello

Could you please share a sample image in order to reproduce the classification process?

Thanks in advance

omaghsoudi commented 2 years ago

You can download some sample images from LIBRA software. In example images of LIBRA, you can find two raw Hologic images.

anlarro commented 2 years ago

Hi. I've tested Deep-LIBRA using the sample images that comes with the LIBRA software. but the results are not good. This is the code I executed python3 Path_to_code/execute_libra_code.py -i ${Path_to_Images} -o ${Path_to_Output} -m ${Path_to_Networks} -ng 1

For the Networks paths I used the folder provided in this link

For cases 2 and 4 the breast_density is successfully estimated but for cases 1,3, and 5 the results are far from satisfactory. This is case 3 (breast density mask was not estimated and the breast mask is not good):

drawing

I've also tested it with my own images but the results are also not good, the breast_density segmentation is not estimated.

Edit: As read in the paper I understand that this behaviour is because Deep-LIBRA was trained using Hologic images, therefore it only works with cases 2 and 4 as these are from Hologic. Any plans to update the trained models for other manufacturers? Or is it possible to provide the code to train the models with our own images? Thank you.

waltman commented 2 years ago

The issue with Hologic vs other manufacturers is almost certainly causing the problem you're seeing.

We've had a group of students updating the Deep LIBRA code as part of their senior design project. Part of that has been rewriting the part of the system that builds the models. We have a meeting with them on Friday and I'll check on when we might be able to release that.

efharkness commented 1 year ago

hello, @waltman please can you advise whether the code has been updated to work with other manufacturers e.g. GE? Many thanks, Elaine