Welcome to the Allen Cell Discussion Forum, the official community site for Allen Institute for Cell Science! This forum is a place for learning, helping, and sharing experiences of using our publicly-available cell lines, plasmids, genomic data, software, and analytic tools.
Get started: read our new user guide and know our community guidelines.
I found the workflows that you share very nicely on your website.
I would like to do segmentation on cells (3d and 2d) mainly to segment the nucleus (DAPI), FISH signal (3d), 3d dots and membranes.
Looking at the # Lookup table of classic image segmentation workflows for 28 intracellular structure localization patterns#
I can easily find which one corresponds to the membrane and the 3d spots BUT I cant find which one is relevant for nucleus segmentation using DAPI (2d / 3d).
My workflow will be :
segmentation of the nucleus (2d / 3d)
Segmentation of the dots 2d / 3d.
hence the first step is to segment the nucleus,
Could you provide me some info or point me into the right direction ?
Thanks for your interest. For nucleus segmentation, we used a method combining training assay and iterative deep learning workflow. You can find more details in our Segmenter paper: https://www.biorxiv.org/content/10.1101/491035v2.full.pdf+html , Section “3D instance segmentation of cell and nucleus”.
The core is the iterative deep learning workflow, which requries preliminary segmentation to start with. In our paper, we use the Training assay approach, where we segmenter lamin B1 (from same multi-channel images of the same cells) and fill the shells segmented from lamin B1 to serve as the groudn truth for nuclei.
Besides training assay, we have seen two other options that could work well to obtain some preliminary segmentations.
(1) for images of high resolution, you may use our released trained models for nuclei segmentation which could give you good results after proper image normalization on your data. Here is our model zoo repo:
if you are interested in this method, I can help work up one demo to show you how to apply our model on your data “after proper intensity normalization”.
(2) for images of low resolution, you may try 2D/3D spot filter with larger kernel size. We have seen successful examples using 2D/3D spot filter generating good preliminary segmentation to start the iterative deep learning workflow.
(1) for images of high resolution, you may use our released trained models for nuclei segmentation which could give you good results after proper image normalization on your data. Here is our model zoo repo:
Regarding your message “if you are interested in this method, I can help work up one demo to show you how to apply our model on your data “after proper intensity normalization”.”
yes please ! that would be great,
I don’t know how you qualify hig res /low res but I am using a Leica 63x with a Leica DFC7000 GT camera…
Please let me know how to plan a demo,
I appreciate your support, this is awesome.
Thanks
antho
I will try to prepare a demo to explain how to apply a released model on new data after propoer normalization, either this Thursday or early next week,