imimd / vessel3ddl

Automated Multiscale 3D Feature Learning for Vessels Segmentation in Thorax CT Images

Vessel3DDL

Automated Multiscale 3D Feature Learning for Vessels Segmentation in Thorax CT Images

Data

The VESSEL 12 data may be downloaded from: https://grand-challenge.org/site/vessel12/ and should be stored at ./Data/VESSEL12/

├── Data
│    └── VESSEL12
│        ├── VESSEL12_01-05
│        ├── VESSEL12_01-20_Lungmasks
│        ├── VESSEL12_06-10
│        ├── VESSEL12_11-15
│        ├── VESSEL12_16-20
│        └── VESSEL12_ExampleScans
│            ├── Annotations
│            ├── Lungmasks
│            └── Scans
├── LICENSE
├── README.md
└── scripts
    ├── config.py
    ├── config.pyc
    ├── LearnClassifier
    ├── LearnDictionary
    ├── UseClassifier
    └── utils

Structure

The entire processing pipeline for the VESSEL12 data is set up in the config.py file.

• Dictionary learning (Unsupervised step). First the dictionary has to be learned on a number of given volumes. The volumes don't have to be annotated.
• Classifier learning (Supervised step). Based on the learned features, train the classifier of choice.
• Testing module. Apply filters from the dictionary and use a classifier.
• Some additional functionality: 3d patch extraction, 3d Gaussian pyramids, loading/saving data. The dictionaries and classifier weights are serialized in the ./Data/Serialized directory.

LearnDictionary

Execute the scripts in following order:

1. ExtractPatches.py
   
2. LearnDictionary.py
   

LearnClassifier

Execute the scripts in following order:

1. ExtractXy_multithread.py
   
2. ConcatenateXy.py
   
3. TrainClassifier.py or MakeMeasurements.py
   

Usage

Once the dictionary and classifier are learned, they can by uses on a given volume.
Execute the scripts in following order:

1. UseClassifier.py
   
2. ViewResults.py
   

Citation

Please cite these paper in your publications if it helps your research:

@inproceedings{konopczynski2019automated,
    title={Automated multiscale 3D feature learning for vessels segmentation in Thorax CT images},
    author={Konopczy{\'n}ski, Tomasz and Kr{\"o}ger, Thorben and Zheng, Lei and Garbe, Christoph S and Hesser, J{\"u}rgen},
    booktitle={2016 IEEE Nuclear Science Symposium, Medical Imaging Conference and Room-Temperature Semiconductor Detector Workshop (NSS/MIC/RTSD)},
    pages={1--3},
    year={2019},
    organization={IEEE}
}

Link to the paper:

• Automated Multiscale 3D Feature Learning for Vessels Segmentation in Thorax CT Images