/MouseCTSegmentation

A deep neural network method for segmenting torso organs from mouse micro-CT images

Primary LanguageJupyter Notebook

MouseCTSegmentation

Introduction

This repository is used for segmenting torso organs from mouse micro-CT images.

Installation

3D-CAFFE

Please follow the instructions to install 3D-CAFFE.

Notes: Don't forget to build Pycaffe interfaces.

make pycaffe && make install # after modify the config file of CAFFE

Python dependencies

The code is developed using python 2.7. Python dependencies are required as follow.

  1. SimpleITK

    pip install SimpleITK

  2. NumPy

    pip install numpy

  3. H5Py

    pip install h5py

Data preparation

  1. CAFFE reads the h5 file to train the network. You need to convert origin formats (e.g. .nii/.nii.gz/.mhd, etc.) to h5 files. We provide a python scriptcov_format.py to convert the format of the files.

  2. You need to create a file name list of the dataset named data.list. It should look like this:

    ./id1.h5
./id2.h5
./id3.h5
...

Usage

  1. Put the dataset in the folder.
  2. Modify the data.list path in the file train.prototxt.
  3. Train the model.
cd MouseCTSegmentation
sh train.sh

Notes: You can specify the numbers of GPU runs in the file train.sh.

  1. Segment the images using the model. We also provide pre-trained models in the folder model. You can modify the variable caffe_model in the file seg.py to use them.
python seg.py

Contact

Please contact hanascend@foxmail.com if you have any questions.