tiaoguoyunduan/kits19

The official repository of the 2019 Kidney and Kidney Tumor Segmentation Challenge

KiTS19

The official 2019 KiTS Challenge repository.

Usage

To get the data for this challenge, please clone this repository (~500MB), and then run get_imaging.py. For example

git clone https://github.com/neheller/kits19
cd kits19
pip3 install -r requirements.txt
python3 -m starter_code.get_imaging

This will download the much larger and static image files from a separate source. The data/ directory should then be structured as follows

data
├── case_00000
|   ├── imaging.nii.gz
|   └── segmentation.nii.gz
├── case_00001
|   ├── imaging.nii.gz
|   └── segmentation.nii.gz
...
├── case_00209
|   ├── imaging.nii.gz
|   └── segmentation.nii.gz
└── kits.json

We've provided some basic Python scripts in starter_code/ for loading and/or visualizing the data.

Loading Data

from starter_code.utils import load_case

volume, segmentation = load_case("case_00123")
# or
volume, segmentation = load_case(123)

Will give you two Nifty1Images. Their shapes will be (num_slices, height, width), and their pixel datatypes will be np.float32 and np.uint8 respectively. In the segmentation, a value of 0 represents background, 1 represents kidney, and 2 represents tumor.

For information about using a Nifty1Image, see the Nibabel Documentation (Getting Started)

Visualizing Data

The visualize.py file will dump a series of PNG files depicting a case's imaging with the segmentation label overlayed. By default, red represents kidney and blue represents tumor.

From Bash:

python3 starter_code/visualize.py -c case_00123 -d <destination>
# or
python3 starter_code/visualize.py -c 123 -d <destination>

From Python:

from starter_code.visualize import visualize

visualize("case_00123", <destination (str)>)
# or
visualize(123, <destination (str)>)

Voxel Spacing

Each Nift1Image object has an attribute called affine. This is a 4x4 matrix, and in our case, it takes the value

array([[0.                          , 0.                      , -1*captured_pixel_width , 0. ],
       [0.                          , -1*captured_pixel_width , 0.                      , 0. ],
       [-1*captured_slice_thickness , 0.                      , 0.                      , 0. ],
       [0.                          , 0.                      , 0.                      , 1. ]])

This information is also available in data/kits.json. Since this data was collected during routine clinical practice from many centers, these values vary quite a bit.

Since spatially inconsistent data might not be ideal for machine learning applications, we have created a branch called interpolated with the same data but with the same affine transformation for each patient.

array([[ 0.        ,  0.        , -0.78162497,  0.        ],
       [ 0.        , -0.78162497,  0.        ,  0.        ],
       [-3.        ,  0.        ,  0.        ,  0.        ],
       [ 0.        ,  0.        ,  0.        ,  1.        ]])

Labeling Errors

We've gone to great lengths to produce the best segmentation labels that we could. That said, we're certainly not perfect. In an attempt to strike a balance between quality and stability, we've decided on the following policy:

If you find an problem with the data, please submit an issue describing it.

Reference

If this data is useful to your research, please cite the following manuscript

@misc{1904.00445,
Author = {Nicholas Heller and Niranjan Sathianathen and Arveen Kalapara and Edward Walczak and Keenan Moore and Heather Kaluzniak and Joel Rosenberg and Paul Blake and Zachary Rengel and Makinna Oestreich and Joshua Dean and Michael Tradewell and Aneri Shah and Resha Tejpaul and Zachary Edgerton and Matthew Peterson and Shaneabbas Raza and Subodh Regmi and Nikolaos Papanikolopoulos and Christopher Weight},
Title = {The KiTS19 Challenge Data: 300 Kidney Tumor Cases with Clinical Context, CT Semantic Segmentations, and Surgical Outcomes},
Year = {2019},
Eprint = {arXiv:1904.00445},
}