Small-organ-segmentation: A Python repository from vanya2v

Small organ segmentation

Small organ segmentation in whole-body scans is challenging. A coarse-to-fine, hierarchical strategy is a common approach to alleviate this problem, with weighting schemes based on auto-context and spatial atlas priors to deal with anatomical variation and class imbalance. This repository is build based on DLTK for FCN baseline and multi-atlas segmentation (http://www.doc.ic.ac.uk/~wbai/software(label_fusion_v2.0.tar.gz) for spatial atlas priors.

Referencing and citing

If you use this multi-modal learning repository in your work please refer to this citation:

Vanya V. Valindria, Ioannis Lavdas, Juan Cerrolaza, Eric O. Aboagye, Andrea G. Rockall, Daniel Rueckert, Ben Glocker. "Small Organ Segmentation in Whole-body MRI using a Two-stage FCN and Weighting Schemes," in 9th International Conference on Machine Learning in Medical Imaging (MLMI), 2018.

Deep Learning Toolkit (DLTK) for Medical Imaging

Setup a virtual environment and activate it. If you intend to run this on machines with different system versions, use the --always-copy flag:
```
virtualenv -p python3 --always-copy venv_tf
source venv_tf/bin/activate
```
Install TensorFlow (>=1.4.0) (preferred: with GPU support) for your system as described here:
```
pip install tensorflow-gpu>=1.4.0
```
Install DLTK: There are two installation options available: You can simply install dltk as is from pypi via
```
pip install dltk
```

Start playing

Coarse-scale segmentation for multi-organ segmentation (coarse_segm) First stage segmentation for multi-class segmentation with weighting schemes on FCN-based segmentation.

 reader_sampling.py    : set the weights per class for imbalance sampling, set each weight relative to the size/proportion of the structure in the volumes
 train_fcn_weighted.py : train the weighted FCN for coarse segmentation
 deploy_save_pmap.py   : get the coarse multi-organ segmentation results and their probability maps for each class

Multi-atlas Prior (MA_prior) Create spatial prior based on multi-organ multi-atlas segmentation.

create_folder.py      : create folder for atlas target and warped images & labels
register.py           : please refer to MRITK software (```https://github.com/BioMedIA/MIRTK```) to register images
transform_label.py    : create warp images and labels
script/labelfusion.py: multi-organ segmentation using PBAF label fusion strategy, and get the spatial prior (save the probabilty maps per class)

Fine-scale segmentation of small organ segmentation (fine_segm) Second stage segmentation for binary small organ segmentation. The cropped ROI input are the images from multiplication of auto-context (probability maps from coarse-scale) and spatial prior (probability maps from multi-atlas segmentation), combined with auto-context (probability maps of the organ from coarse-scale segmentation.

 reader_bladder.py     : reader for binary small organ (here the organ is bladder) segmentation. Set the weight high for foreground class.
 train_fcn_bladder.py : train the weighted FCN for bladder segmentation. Input: cropped ROI (multiplication with spatial prior) of images and probabilty maps of bladder segmentation (from coarse-scale). Set the example size small (8x8x8) as the input size now is cropped (smaller).
 deploy_bladder.py   : get the binary small-organ segmentation result

License

See LICENSE

vanya2v/Small-organ-segmentation