[3dem] MemBrain automated membrane segmentation: beta release, please try!

[Ben Engel]

Hello Cryo-ET world!

We are currently developing *MemBrain**-seg*, our new deep
learning-based membrane
segmentation tool, and we would like to get your feedback!

So far, we trained a neural network on a dataset of diverse membrane
architectures, and it predicts membrane segmentations quite well on our
data:

[image: image.png]

Now, we would like to test the network's generalizability on tomograms from
around the TeamTomo-verse. We would highly appreciate if you could give
MemBrain-seg a try on your data and let us know how it works for you. We
aim to make using MemBrain-seg as effortless as possible: with its
user-friendly
command line interface, you can segment your tomograms within minutes.

[image: image.png]

You can access MemBrain-seg on the *#teamtomo Github repository*:
https://urldefense.com/v3/__https://github.com/teamtomo/membrain-seg__;!!Mih3wA!H5hsdNPpe5GAI-V-hmcDyqudo1Lc8qoeJ-6cy6VLxyyPCovRat0eKWRS909vXDWQDkfGHK-fyZNr81VLWNSNORI$ 

The package includes functionalities for segmenting your tomograms, some
optional preprocessing steps, as well as tools for retraining the model
with your own data to fine-tune the performance (although MemBrain-seg is
intended to function well "out-of-the-box" without retraining). For all
these functionalities, you can find more information in our documentation (
https://urldefense.com/v3/__https://github.com/teamtomo/membrain-seg/blob/main/docs/index.md__;!!Mih3wA!H5hsdNPpe5GAI-V-hmcDyqudo1Lc8qoeJ-6cy6VLxyyPCovRat0eKWRS909vXDWQDkfGHK-fyZNr81VLFwwqzEo$ ), or even
in our YouTube video series about creating new training patches from your
tomograms (
https://urldefense.com/v3/__https://www.youtube.com/playlist?list=PLV3O3yHyCjkXAi9MWgComzh6JuKcHUgNU__;!!Mih3wA!H5hsdNPpe5GAI-V-hmcDyqudo1Lc8qoeJ-6cy6VLxyyPCovRat0eKWRS909vXDWQDkfGHK-fyZNr81VLx29hkX4$ ).

MemBrain-seg is still under early development and may not work perfectly on
your data. In this case, we recommend to extract more patches from your
tomograms and include them in the training dataset. In our preliminary
testing, performance improves significantly after only correcting a few new
patches. We are eager to extend the MemBrain-seg training dataset to cover
more and more different views of membranes from a variety of cells and tomo
acquisition parameters.

So please reach out to us if you would like to collaborate on extending
MemBrain-seg dataset to help improve the general performance for everyone.
It is not necessary to share full tomograms, only the small corrected
patches. If you have any questions or trouble running MemBrain-seg, please
also let us know – we are happy to receive feedback and assist in getting
MemBrain-seg running well for you. We're looking forward to hearing about
your experiences!

MemBrain-seg is the product of a great collaboration with Alister Burt (MRC
LMB / Genentech) and Kevin Yamauchi (ETH Zürich).


Best wishes,
Lorenz, Simon, Ben, Tingying, and the rest of the MemBrain team
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20230704/97db9346/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 218309 bytes
Desc: not available
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20230704/97db9346/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 1248273 bytes
Desc: not available
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20230704/97db9346/attachment-0003.png>