DeepSlice is a python library which automatically aligns mouse histology with the allen brain atlas common coordinate framework (and now rat brain histology to the Waxholm rat brain atlas, though this is in beta).
The alignments are viewable, and refinable, using the QuickNII software package.
DeepSlice requires no preprocessing and works on any stain, however we have found it performs best on brightfield images.
At present one limitation is that it only works on Coronally cut sections, we will release an update in the future for sagittal and horizontally cut histology.
DeepSlice automates the process of identifying exactly where in the brain a section lies, it can accomodate non-orthogonal cutting planes and will produce an image specific annotation for each section in your brain.
If you use DeepSlice in your work please cite Carey et al, 2023. It may also be useful if you mention the version you use :)
DeepSlice is fully integrated with the QUINT workflow. Quint helps you register, segment and quantify brain wide datasets! Β ππ§ π¬π»π€
If you would like to use DeepSlice but don't need your own personal installation, check out DeepSlice Flask, a fully functional web application which will allow you to upload your dataset and download the aligned results. The web interface was developed by Michael Pegios.
Happy Aligning :)
This is the easy and recommended way to install DeepSlice, first make sure you have Python 3.11 installed and then simply:
pip install DeepSliceAnd you're ready to go! π Check out the PyPi package here
If you run into any problems create a github issue and I will help you solve it.
After cloning our repo and navigating into the directory open an ipython session and import our package.
from DeepSlice import DSModel Next, specify the species you would like to use and initiate the model.
species = 'mouse' #available species are 'mouse' and 'rat'
Model = DSModel(species)Important
-
Sections in a folder must all be from the same brain
-
DeepSlice uses all the sections you select to inform its prediction of section angle. Thus it is important that you do not include sections which lie outside of the Allen Brain Atlas. This include extremely rostral olfactory bulb and caudal medulla. If you include these sections in your selected folder it will reduce the quality of all the predictions.
-
The sections do not need to be in any kind of order.
-
The model downsamples images to 299x299, you do not need to worry about this but be aware that there is no benefit from using higher resolutions.
Now your model is ready to use, just direct it towards the folder containing the images you would like to align.
eg:
βββ your_brain_folder
β βββ brain_slice_1.png
β βββ brain_slice_2.png
β βββ brain_slice_3.pngIn this parent directory there should be only one sub folder, in this example this is "your_brain_folder".
To align these images using DeepSlice simply call
folderpath = 'examples/example_brain/GLTa/'
#here you run the model on your folder
#try with and without ensemble to find the model which best works for you
#if you have section numbers included in the filename as _sXXX specify this :)
Model.predict(folderpath, ensemble=True, section_numbers=True)
#If you would like to normalise the angles (you should)
Model.propagate_angles()
#To reorder your sections according to the section numbers
Model.enforce_index_order()
#alternatively if you know the precise spacing (ie; 1, 2, 4, indicates that section 3 has been left out of the series) Then you can use
#Furthermore if you know the exact section thickness in microns this can be included instead of None
#if your sections are numbered rostral to caudal you will need to specify a negative section_thickness
Model.enforce_index_spacing(section_thickness = None)
#now we save which will produce a json file which can be placed in the same directory as your images and then opened with QuickNII.
Model.save_predictions(folderpath + 'MyResults')
We are grateful to Ann Goodchild for her time-saving blunt assessments of many failed prototypes, for the motivation provided by Dr William Redmond, and especially to Veronica Downs, Freja Warner Van Dijk and Jayme McCutcheon, whose Novice alignments were instrumental to this work. We would like to thank Gergely CsΓΊcs for providing his expertise and many atlasing tools. Work in the authorsβ laboratories is supported by the National Health & Medical Research Council of Australia, the Hillcrest Foundation, and Macquarie University (SMcM), and from the European Unionβs Horizon 2020 Framework Program for Research and Innovation under the Specific Grant Agreement No. 945539 (Human Brain Project SGA3) and the Research Council of Norway under Grant Agreement No. 269774 (INCF, JGB). We are grateful to Macquarie University for access to their HPC resources, essential for production of early DeepSlice prototypes.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent