Implementation of the paper Using Simulated Training Data of Voxel-level Generative Model to Improve 3D Neuron Reconstruction.

In this paper, we propose a novel strategy of using two-stage generative models to simulate training data with voxel-level labels (MPGAN). Trained upon unlabeled data by optimizing a novel objective function of preserving predefined labels, the models are able to synthesize realistic 3D images with underlying voxel labels. The framework of MPGAN model is shown as follow:

• Linux
• NVIDIA GPU
• TensorFlow==1.15.0

• Clone this repo:

git clone https://github.com/CSDLLab/MPGAN.git
cd MPGAN

• Create virtual environment and install required packages:

conda create -n mpgan
conda activate mpgan
conda install --yes --file requirements.txt

• Use first-stage simulator to generate the fake images:

python 1_simulator.py

• The output images are shown in follow figure:

• Use second-stage MPGAN to refine the fake images:

python 2_mpgan.py

Due to the size of each individual stack, which can be as big as 512 * 512 * 512, it is not practical to input the whole stack into a deep neural network. Instead, we use small patches with the size of 64 * 64 * 32. To exclude patches containing too few foreground voxels, we use a sliding window to traverse each stack.

• Use OTSU to segment the image.

• Skeletonize the neuron image to get the neuron centerline.

• Slide the window to traverse the image on centerline to ensure that the neuron branches are centered in the training data.

If you find MPGAN useful in your work, please consider citing our paper:

@article{liu2022using,
  title={Using simulated training data of voxel-level generative models to improve 3D neuron reconstruction},
  author={Liu, Chao and Wang, Deli and Zhang, Han and Wu, Wei and Sun, Wenzhi and Zhao, Ting and Zheng, Nenggan},
  journal={IEEE transactions on medical imaging},
  volume={41},
  number={12},
  pages={3624--3635},
  year={2022},
  publisher={IEEE}
}