MEMOS: New extension for deep-learning based segmentation of 3D fetal mice scans.

Announcements
1

We would like to announce a new Slicer extension, MEMOS, for deep-learning based segmentation of diceCT scans of fetal mice. See the repository for quick installation instructions.

Accompanying paper can be found in Biology Open: Deep learning enabled multi-organ segmentation of mouse embryos | Biology Open | The Company of Biologists

Using GPU, MEMOs provide segmentation of 50 anatomical structures in the E15.5 atlas provided by the International Mouse Phenotyping Consortium in about 40-60 seconds.

In future, we plan to incorporate more developmental time points as a separate trained networks.

image
image1280×932 111 KB

12 Likes
2

Hi, that sounds great, I’d like to try it, but I’m unable to install this extension
image

3

Which os do you use?

That might be related with the patch release last night. I just downloaded and install fresh on windows (v5.2.2), can you try again?

4

I’m using WIN10 and slicer 5.2.1 r31317 / 77da381, still not working, I’ll update to 5.2.2 and try again. Thank you

5

İ see. As it is a new extension, it is only available for stable or 5.3.0 (or higher) previews.
Yes, try with latest stable.

6

I have only used Total Segmentator so far but MEMOS also looks fantastic.

Is there anything comparable available for adult mice? I have >1000 CT scans of mice that need to be segmented (mainly liver, kidneys, and heart).

Best, Jan

7

No, we don’t have anything for adult mice. We don’t really do physiology, mostly development.

However, it is getting easier to train deep learning networks for custom segmentation tasks. So if you have already manually segmented ones that can be used as training data, it might be just a matter of putting some GPUs to use.