Skip to content

Latest commit

 

History

History
113 lines (110 loc) · 4.76 KB

README.md

File metadata and controls

113 lines (110 loc) · 4.76 KB
Raw

3DUNet implemented with pytorch

Introduction

The repository is a 3DUNet implemented with pytorch, referring to this project. I have redesigned the code structure and used the model to perform liver and tumor segmentation on the lits2017 dataset.
paper: 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation

Requirements:

pytorch >= 1.1.0
torchvision
SimpleITK
Tensorboard
Scipy

Code Structure

├── dataset          # Training and testing dataset
│   ├── dataset_lits_train.py 
│   ├── dataset_lits_val.py
│   ├── dataset_lits_test.py
│   └── transforms.py 
├── models           # Model design
│   ├── nn
│   │   └── module.py
│   │── ResUNet.py      # 3DUNet class
│   │── Unet.py      # 3DUNet class
│   │── SegNet.py      # 3DUNet class
│   └── KiUNet.py      # 3DUNet class
├── experiments           # Trained model
|── utils            # Some related tools
|   ├── common.py
|   ├── weights_init.py
|   ├── logger.py
|   ├── metrics.py
|   └── loss.py
├── preprocess_LiTS.py  # preprocessing for  raw data
├── test.py          # Test code
├── train.py         # Standard training code
└── config.py        # Configuration information for training and testing

Quickly Start

1) LITS2017 dataset preprocessing:

  1. Download dataset from google drive: Liver Tumor Segmentation Challenge.
    Or from my share: https://pan.baidu.com/s/1WgP2Ttxn_CV-yRT4UyqHWw Extraction code：hfl8 (The dataset consists of two parts: batch1 and batch2)
  2. Then you need decompress and merge batch1 and batch2 into one folder. It is recommended to use 20 samples(27~46) of the LiTS dataset as the testset and 111 samples(0~26 and 47~131) as the trainset. Please put the volume data and segmentation labels of trainset and testset into different local folders, such as:
raw_dataset:
    ├── test  # 20 samples(27~46) 
    │   ├── ct
    │   │   ├── volume-27.nii
    │   │   ├── volume-28.nii
    |   |   |—— ...
    │   └── label
    │       ├── segmentation-27.nii
    │       ├── segmentation-28.nii
    |       |—— ...
    │       
    ├── train # 111 samples(0\~26 and 47\~131)
    │   ├── ct
    │   │   ├── volume-0.nii
    │   │   ├── volume-1.nii
    |   |   |—— ...
    │   └── label
    │       ├── segmentation-0.nii
    │       ├── segmentation-1.nii
    |       |—— ...
  1. Finally, you need to change the root path of the volume data and segmentation labels in ./preprocess_LiTS.py, such as:
    row_dataset_path = './raw_dataset/train/'  # path of origin dataset
    fixed_dataset_path = './fixed_data/'  # path of fixed(preprocessed) dataset
  1. Run python ./preprocess_LiTS.py
    If nothing goes wrong, you can see the following files in the dir ./fixed_data
│—— train_path_list.txt
│—— val_path_list.txt
│
|—— ct
│       volume-0.nii
│       volume-1.nii
│       volume-2.nii
│       ...
└─ label
        segmentation-0.nii
        segmentation-1.nii
        segmentation-2.nii
        ...

2) Training 3DUNet

  1. Firstly, you should change the some parameters in config.py,especially, please set --dataset_path to ./fixed_data
    All parameters are commented in the file config.py.
  2. Secondely,run python train.py --save model_name
  3. Besides, you can observe the dice and loss during the training process in the browser through tensorboard --logdir ./output/model_name.

3) Testing 3DUNet

run test.py
Please pay attention to path of trained model in test.py.
(Since the calculation of the 3D convolution operation is too large, I use a sliding window to block the input tensor before prediction, and then stitch the results to get the final result. The size of the sliding window can be set by yourself in config.py)

After the test, you can get the test results in the corresponding folder:./experiments/model_name/result

You can also read my Chinese introduction about this 3DUNet project here. However, I no longer update the blog, I will try my best to update the github code.
If you have any suggestions or questions, welcome to open an issue to communicate with me.