Skip to content

Latest commit

 

History

History
115 lines (82 loc) · 7.53 KB

README.md

File metadata and controls

115 lines (82 loc) · 7.53 KB
Raw

DOI PyPI - Version PyPI - Downloads Pepy Total Downloads

Alt 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. Alt 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.

Citation

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 :)

Workflow

DeepSlice is fully integrated with the QUINT workflow. Quint helps you register, segment and quantify brain wide datasets!   🐭🧠🔬💻🤖

Web Application

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.

Installation: How to install DeepSlice

Usage: How to align using DeepSlice

For a jupyter notebook example check out

Happy Aligning :)


Installation

From PIP

This is the easy and recommended way to install DeepSlice, first make sure you have Python 3.11 installed and then simply: 
pip install DeepSlice

And you're ready to go! 🚀 Check out the PyPi package here

If you run into any problems create a github issue and I will help you solve it.


Basic Usage

On start

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.

Predictions

Now your model is ready to use, just direct it towards the folder containing the images you would like to align.

eg:

    
 ├── your_brain_folder
 │   ├── brain_slice_1.png 
 │   ├── brain_slice_2.png     
 │   ├── brain_slice_3.png

In 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')

Acknowledgements

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 Csúcs for providing his expertise and many atlasing tools. Work in the authors’ laboratories is supported by the National Health & Medical Research Council of Australia, the Hillcrest Foundation, and Macquarie University (SMcM), and from the European Union’s 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.