mister-jia / airway-summerschool-2022 Goto Github PK

BAS
└───train
│   │   case1
│   │   case2
│   └───casex
│       │   input
│       │   mask
│       │   processed
│           │   casex_clean.nii
│           │   casex_label.nii
│           └───casex_box.npy
│   
└───test
    │   case1
    │   case2

Use the casex_clean.nii for input CT volumes and the corresponding casex_label.nii for the ground-truth. For space room saving, the input and mask folder may be discarded.

• ITK-SNAP : portable.
• 3D-Slicer : various modules and extensions.

1. First, we provide a Demo Inference with a baseline trained weight file that helps you get familiar with the airway prediction procedure via the CNN output.

   The entrance is at main_code/scripts/Demo_Inference/demo_test.py. You can try the following command:

   python demo_test.py --input_path $INPUT_PATH --output_path $OUTPUT_PATH
   

   You could use the ITK-SNAP / 3D-Slicer or any visualization tools to check your results.

2. Second, we provide a Train Pipeline that helps you train the airway tree modeling task.

   The entrance is at main_code/pipeline/pipeline_train_airway_IMR-Summer-School-2022.py. You can try the following command:

   python pipeline_train_airway_IMR-Summer-School-2022.py --dataroot $DATASET_DIR --name $EXPERIMENT_NAME --checkpoints_dir $MODEL_LOADDIR \ 
   --model $MODEL --dataset_mode $DATASET_MODE --in_channels $INPUT_CH --out_channels $OUTPUT_CH --gpu_ids $GPU_IDS --suffix $SUFFIX  
   

   TIPS: the detailed arguments for base configures and training procedures are in *** main_code/options/base_options*** and main_code/options/train_options respectively. Please refer to these two files and specify the arguments in the default settings or the command line.

3. Third, we provide some ports for you to extend the project. Specifically,

   • Data Augmentation: In main_code/dataloader/airway_dataset.py, you can conduct extra data augmentation.
   • Loss Function: In main_code/util/losses.py, you can construct other loss functions and call it in the *** main_code/models/unet3d_model.py***
   • Model Design: In main_code/models/ you can design your own models inherited from the base_model and use the modules in the main_code/models/networks.py

Python >= 3.8. The deep learning framework is PyTorch=1.11.0 and Torchvision = 0.12.0

Some python libraries are also necessary, you can use the following command to set up.

pip install -r requirements.txt