Code

CShaper is developed with Tensforflow 1.12 and python3.6 on a Linux workstation (Intel Xeon Silver 4116 CPU 2.10GHz x 48 with 125GB memory). The source code of CShaper can be found at our Github repository. To facilitate the deployment of CShaper, we also provide Dockerfile to build a docker container (Dockerfile). Because the CShaper image is built based on public resource deepo and nvidia-docker, please make sure you have installed all requirements before run docker build -t yourname/CShaper. Steps to use our docker are listed as follows:

1. Download training or testing data to folder Data. The data should be saved like:

   --Data root folder
    |--MembTest: testing data
        |--181210plc1p2: embryo smaple name
            |--RawMemb: 3D membrane image
            |--RawNuc: 3D nucleus image
            |--SegNuc: 3D nucleus segmentation (only nucleus segmentation from AceTree)
            |--CD181210plc1p2.csv: nucleus lineage file manually corrected from AceTree and StarryNite.
        |--...
    |--MembTraining: training data
        |--170704plc1p2: embryo sample with annotations
            |--RawMemb: 3D membrane image
            |--RawNuc: 3d nucleus image
            |--SegNuc: 3D nucleus segmentation
            |--Segmemb: 3D cell-level annotations
            |--CD170704plc1p2.csv: nucleus lineage
        |--...
   

2. Build CShaper image with Dockerfile (change yourname to your Docker account):

   $ docker build -t yourname/CShaper
   

3. Run CShaper container:

   $ docker run -rm -it -v /absolute/path/to/Data:/home/CShaper/Data -v /absolute/path/to/ResultCell:/home/CShaper/Data \ 
   -v /absolute/path/to/RobustStat:/home/CShaper/ShapeUtil/RobustStat yourname/CShaper
   $ cd /home/CShaper
   

   Data and result folders are shared between the container and host in order to keep the processed files in the host.

4. After building the container, users can refer to our code repository to train or test CShaper.