mrid-python 0.1.4

Tools for working with 3D medical images and segmentations - registration, brain skull-stripping, etc.

mrid

mrid is a library for preprocessing of 3D images, particularly medical images.

It provide interfaces for many medical image processing tools such as SimpleElastix, HD-BET, SynthStrip, CTSeg. Note that those libraries are not bundled with mrid, I've included installation instructions in all notebooks.

Installation

Either run

pip install mrid-python

or

pip install git+https://github.com/inikishev/mrid

Basics

The images you pass to all functions in mrid can be path to a .nii.gz file, DICOM directory, sitk.Image, numpy array or torch tensor. All functions return results as sitk.Image. If you need a numpy array, you can use mrid.tonumpy(sitk_image).

Registering images with SimpleITK-SimpleElastix

SimpleElastix is a robust tool for image registration which works really well out-of-the-box. It works on both Windows and Linux.

See this notebook for how to install and use it.

Skullstripping MRI scans with HD-BET

HD-BET is a model that performs skullstripping of pre- and post-constrast T1, T2 and FALIR MRIs. It works on both Windows and Linux.

See this notebook for how to install and use it

Skullstripping with SynthStrip

SynthStrip is a skull-stripping tool that works with many different image types and modalities, including MRI, DWI, CT, PET, etc.

See this notebook for how to install and use it

Skullstripping and segmentation of CT images with CTseg

CTseg can skull-strip CT images and perform their segmentation, it also registers them to a common space (see its README). Note that it can be very slow for 512x512 series (can take few hours), but you can downsample to 256x256. If you only need to quickly skullstrip CT scans without warping them you can use SynthStrip.

TODO！！！

Example workflow - preprocessing MRIs to BraTS format

Many BraTS datasets are provided as skullstripped images in SRI24 space. See this notebook for how to process raw scans to this format.

(T1n image looks weird because that's just how it is in the dataset)

References

The MRIs for all images above are from https://zenodo.org/records/7213153.

Colin Vanden Bulcke. (2022). Open-Access DICOM MRI session (1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7213153