3D-Registration tool for H&E mouse brain coronal sections

Hello, Image Analysis community!

Background

Along with my colleagues, we took images of several brain coronal sections after H&E free-floating staining (species: mouse).
For image acquisition, we used a 20X objective of the Digital whole slide scanner “Nanozoomer XR” (Hamamatsu, Photonics Ltd.) thus we have imaged each coronal section entirely, and they look like these:

Analysis goals

We are interested in measuring the corpus callosum thickness as well as the corpus callosum volume.
We’re currently using the wonderful QuPath software to measure corpus callosum thickness manually on each brain slice, simply by taking a ruler and measuring the space encompassing the structure.

Challenges

I am aware this approach has several limitations, in terms of sensitivity of our measures and other factors such as not being absolutely sure that we’re on the exact same Z-planes all the time across our samples and our measures.

Here’s what I’d like to know:

I was wondering if having a 3D-Registration tool could allow us to add several coronal sections of the same sample together, register them in 3D, and finally reconstitute with a fair approximation the 3D volume of the corpus callosum.

I thought this could give us more information from our data; increasing also the sensitivity of our measures that otherwise would just be limited to our “ruler” measurements that I mentioned before.

I’d be very interested to know from you whether you think there’s any tool available that could tackle some of the aspects I mentioned.
Please remember that we’re working specifically with H&E staining mouse coronal sections imaged with a whole-slide scanner (the final format is .ndpi -which can be opened in QuPath).

Thanks a lot everyone,
looking forward to hearing your thoughts!

1 Like

You could try using QuPath Alignment of Serial Images (QuASI) GitHub - MarkZaidi/QuASI: List of QuPath scripts for alignment and stain deconvolution of whole-slide histology images. It’s main goal is to align (and optionally perform stain deconvolution) of serial sections. The output will be a multichannel OME-tiff, where each channel corresponds to the R, G, or B values of an individual slice.

To turn this into a 3D volume, you’d need to do this in an external programming language such as MATLAB or Python. Take the output of QuASI, load it in, and transform every 3 channels (RGB) of the aligned image into a Z stack with an offset corresponding to the tissue depth + tissue thickness.

Alternatively, there was a paper a while back where they took thousands of serial sections of a tumour, stained for H&E, and used an automatic approach to align and generate a 3D volume https://www.biorxiv.org/content/10.1101/2020.12.08.416909v1. The tool they built to achieve this is CODA, might be worthwhile looking in to if you’re working with more than a dozen serial sections.

1 Like