I coded a personal pipeline for pixelwise registration of brain regions and decided to package it for broader use. Based off of the vedo and brainrender frameworks, I’ve finished the first tentative version of brainreg3D. This is a python package for the manual registration of 2D images over a 3D brain. Below is the description as found at https://github.com/JoeRicotta/brainreg3D/blob/master/description.md.
brainreg3D is based off of the vedo and brainrender framework to allow registration of a 2D image onto a 3D brain. It uses perspectival projections of the brain onto the field of view to map cortical regions to image pixels.
The following images were obtained with the below code:
from brainreg3D import BrainReg3D
# by specifying the image_dims_mm parameter, we ensure the loaded image
# is to scale
reg = BrainReg3D('./resources/sample_image.tif', image_dims_mm=[6.25,4])
# if verbose, intermediate processing steps will be plotted and shown
reg.verbose = True
# run the pipeline
reg.run()
The first step is to manually align the image over the Allen atlas.
After hitting enter, an image with the perspectival projections of the underlying brain regions will appear. These brain regions can be identified (and modified) within the class parameter BrainReg3D._DEFAULT_BRAIN_REGIONS.
The underlying volumes of these specified regions are then plotted:
The 2D projections of these regions are shown, without regard for their depth in the brain.
Superficial structures mask deep structures. To account for this, the image space is quantized and projections are preserved based on the length of these projections:
A k-nearest neigbors algorithm is then used to label each pixel identity based on the identity of it’s neigbors:
Finally, the pixel labels are converted to masks (images of 0 and 1 values) and saved in the image directory.
