nnunet入门之二 (MRI图像分割)

MIC-DKFZ/nnUNet
选择Linux环境运行该项目，Windows环境需要更改较多的参数，暂不支持。

安装环境

可以查看另一篇

数据处理

因为开源的MRI数据一般是4D的，即把所有模态拼接起来，可以用nnUNet的转换命令；私人数据一般是存放3D的不同模态，可以按照本文私人数据集（3D）的方式处理，也可以将所有模态拼接起来，用nnUNet的转换命令。

1. 开源数据集（4D）

   1.1 文件夹目录

   └─Task01_BrainTumour
      │  dataset.json
      │  
      ├─imagesTr
      │      BRATS_001_0000.nii.gz
      │      BRATS_001_0001.nii.gz
      │      BRATS_001_0002.nii.gz
      │      BRATS_001_0003.nii.gz
      │      BRATS_002_0000.nii.gz
      │      BRATS_002_0001.nii.gz
      │      BRATS_002_0002.nii.gz
      │      BRATS_002_0003.nii.gz
      │      BRATS_003_0000.nii.gz
      │      BRATS_003_0001.nii.gz
      │      BRATS_003_0002.nii.gz
      │      BRATS_003_0003.nii.gz
      │      BRATS_004_0000.nii.gz
      │      BRATS_004_0001.nii.gz
      │      BRATS_004_0002.nii.gz
      │      BRATS_004_0003.nii.gz
      │      BRATS_005_0000.nii.gz
      │      BRATS_005_0001.nii.gz
      │      BRATS_005_0002.nii.gz
      │      BRATS_005_0003.nii.gz
      │      ...
      ├─imagesTs
      │      BRATS_485_0000.nii.gz
      │      BRATS_485_0001.nii.gz
      │      BRATS_485_0002.nii.gz
      │      BRATS_485_0003.nii.gz
      │      BRATS_486_0000.nii.gz
      │      BRATS_486_0001.nii.gz
      │      BRATS_486_0002.nii.gz
      │      BRATS_486_0003.nii.gz
      │      ...
      └─labelsTr
             BRATS_001.nii.gz
             BRATS_002.nii.gz
             BRATS_003.nii.gz
             BRATS_004.nii.gz
             BRATS_005.nii.gz
             ...
   

   1.2 json文件信息

   nnUNet/nnunet/dataset_conversion/utils.py里面的函数generate_dataset_json可以生成相应任务的json文件。

   { 
   "name": "BRATS", 
   "description": "Gliomas segmentation tumour and oedema in on brain images",
   "reference": "https://www.med.upenn.edu/sbia/brats2017.html",
   "licence":"CC-BY-SA 4.0",
   "release":"2.0 04/05/2018",
   "tensorImageSize": "4D",
   "modality": { 
   	 "0": "FLAIR", 
   	 "1": "T1w", 
   	 "2": "t1gd",
   	 "3": "T2w"
    }, 
    "labels": { 
   	 "0": "background", 
   	 "1": "edema",
   	 "2": "non-enhancing tumor",
   	 "3": "enhancing tumour"
    }, 
    "numTraining": 484, 
    "numTest": 266,
    "training":[{"image":"./imagesTr/BRATS_001.nii.gz","label":"./labelsTr/BRATS_001.nii.gz"},{"image":"./imagesTr/BRATS_002.nii.gz","label":"./labelsTr/BRATS_002.nii.gz"},{"image":"./imagesTr/BRATS_003.nii.gz","label":"./labelsTr/BRATS_003.nii.gz"},{"image":"./imagesTr/BRATS_004.nii.gz","label":"./labelsTr/BRATS_004.nii.gz"},{"image":"./imagesTr/BRATS_005.nii.gz","label":"./labelsTr/BRATS_005.nii.gz"},...],
    "test":["./imagesTs/BRATS_485.nii.gz","./imagesTs/BRATS_486.nii.gz",...]
    }
   
   

   1.3 转换数据

   nnUNet_convert_decathlon_task -i /xxx/Task01_BrainTumour
   

   转换的数据存在nnUNet_raw_data_base/nnUNet_raw_data/Task001_BrainTumour，此时imagesTr和imagesTs里的文件名加了后缀"_0000", "_0001", "_0002", "_0003"。

   注意：此处Task01_BrainTumour变为Task001_BrainTumour。

2. 私人数据集（3D）

   直接转换数据，将数据存放到nnUNet_raw_data_base/nnUNet_raw_data/Task001_BrainTumour

   2.1 文件夹目录

   └─Task001_BrainTumour
      │  dataset.json
      │  
      ├─imagesTr
      │      BRATS_001.nii.gz
      │      BRATS_002.nii.gz
      │      BRATS_003.nii.gz
      │      BRATS_004.nii.gz
      │      BRATS_005.nii.gz
      │      ...
      ├─imagesTs
      │      BRATS_485.nii.gz
      │      BRATS_486.nii.gz
      │      ...
      └─labelsTr
             BRATS_001.nii.gz
             BRATS_002.nii.gz
             BRATS_003.nii.gz
             BRATS_004.nii.gz
             BRATS_005.nii.gz
             ...
   

   2.2 json文件信息

   nnUNet/nnunet/dataset_conversion/utils.py里面的函数generate_dataset_json可以生成相应任务的json文件。

   { 
   "name": "BRATS", 
   "description": "Gliomas segmentation tumour and oedema in on brain images",
   "reference": "https://www.med.upenn.edu/sbia/brats2017.html",
   "licence":"CC-BY-SA 4.0",
   "release":"2.0 04/05/2018",
   "tensorImageSize": "4D",
   "modality": { 
   	 "0": "FLAIR", 
   	 "1": "T1w", 
   	 "2": "t1gd",
   	 "3": "T2w"
    }, 
    "labels": { 
   	 "0": "background", 
   	 "1": "edema",
   	 "2": "non-enhancing tumor",
   	 "3": "enhancing tumour"
    }, 
    "numTraining": 484, 
    "numTest": 266,
    "training":[{"image":"./imagesTr/BRATS_001.nii.gz","label":"./labelsTr/BRATS_001.nii.gz"},{"image":"./imagesTr/BRATS_002.nii.gz","label":"./labelsTr/BRATS_002.nii.gz"},{"image":"./imagesTr/BRATS_003.nii.gz","label":"./labelsTr/BRATS_003.nii.gz"},{"image":"./imagesTr/BRATS_004.nii.gz","label":"./labelsTr/BRATS_004.nii.gz"},{"image":"./imagesTr/BRATS_005.nii.gz","label":"./labelsTr/BRATS_005.nii.gz"},...],
    "test":["./imagesTs/BRATS_485.nii.gz","./imagesTs/BRATS_486.nii.gz",...]
    }
   
   

预处理

# 只进行3d预处理，不进行2d预处理
nnUNet_plan_and_preprocess -t 01 -pl2d None

训练

可以查看另一篇

测试

可以查看另一篇