A neural networks toolbox with a focus on medical image analysis in tensorflow/keras for now.
To use the Neurite library, either clone this repository and install the requirements listed in setup.py
or install directly with pip.
pip install neurite
- layers: various network layers, sparse operations (e.g.
SpatiallySparse_Dense
), andLocallyConnected3D
currently not included inkeras
- utils: various utilities, including
interpn
: N-D gridded interpolation, and several nonlinearities - models: flexible models (many parameters to play with) particularly useful in medical image analysis, such as UNet/hourglass model, convolutional encoders and decoders
- generators: generators for medical image volumes and various combinations of volumes, segmentation, categorical and other output
- callbacks: a set of callbacks for
keras
training to help with understanding your fit, such as Dice measurements and volume-segmentation overlaps - dataproc: a set of tools for processing medical imaging data for preparation for training/testing
- metrics: metrics (most of which can be used as loss functions), such as Dice or weighted categorical crossentropy
- plot: plotting tools, mostly for debugging models
- tensorflow, keras and all of their requirements (e.g. hyp5)
- numpy, scipy, tqdm
- pystrum
If you use this code, please cite:
Anatomical Priors in Convolutional Networks for Unsupervised Biomedical Segmentation
Adrian V. Dalca, John Guttag, Mert R. Sabuncu
CVPR 2018.
[ PDF | arxiv | bibtex ]
If you are using any of the sparse/imputation functions, please cite:
Unsupervised Data Imputation via Variational Inference of Deep Subspaces
Adrian V. Dalca, John Guttag, Mert R. Sabuncu
Arxiv preprint 2019
[ arxiv | bibtex ]
Please open an issue [preferred] or contact Adrian Dalca at [email protected] for question related to neurite
.
Parts of neurite
were used in VoxelMorph and brainstorm, which we encourage you to check out!