README.md

Back to Projects List

Automatic Segmentation Using Neural Network

Key Investigators

• Bence Horvath (University of Szeged)
• Attila Nagy (Faculty of Medicine, University of Szeged)
• Kitti Farkas (University of Szeged)
• Endre Vecsernyes (University of Szeged)
• Deepak Chittajallu (Kitware) available remotely ([email protected], @cdeepakroy)

Project Description

We want to implement a neural network based automatic segmentation algorithm to segment the upper region of airways in CT images. We have approximately 40 non segmented CT images of the neck in DICOM files, that we have to segment for the training phase. For medical or biological images is a great solution the U-net, that is a special type of neural networks. To compare with other neural network based algorithm, U-net doesn't need large dataset (~30-40 images enough).

Objective

1. Objective A. Making a training data set (~30-40 images). Find a quickly and precise way to segment the upper airways in 3D Slicer.
2. Objective B. Make a baseline U-net, in python programming language whit this dataset.
3. Objective C. Testing the program.

Approach and Plan

1. Tba.
2. ...
3. ...

Progress and Next Steps

We made some segmentation mask in Slicer, using Segment editor's fill between slices tool.

• Few manually segmented images
• Few segmentation masks

Illustrations

Keras, a neural network package in python:

• https://keras.io/

The U-net:

• https://lmb.informatik.uni-freiburg.de/people/ronneber/u-net/

Background and References

• Source code: https://github.com/YourUser/YourRepository
• Documentation: https://link.to.docs
• Test data: https://link.to.test.data