2d segmentation for LITS2017 (the challenge website in codalab)

How to use it

step 1 data process

python preprocess.py

you can get the data like this,each npy file shape is 448*448*3,use each slice below and above as input image,the mid slice as mask(448*448)

data---
    trainImage_k1_1217---
        1_0.npy
        1_1.npy
        ......
    trainMask_k1_1217---
        1_0.npy
        1_1.npy
        ......

step 2 Train the model

python train.py

step 3 Test the model

you can download the models and ROI liver mask from This(syzv) and put them in suitable place according to test.py python test.py

step 4 postprocess to remove False predict

python postprocess.py

baseline

Method	U-Net	Att U-Net	sep U-Net	denseunet
`Dice(liver)`	0.951	0.950	0.948	0.949
`rvd`	0.016	0.038	0.037	0.029
`jaccard`	0.911	0.906	0.903	0.904
`Dice(tumor)`	0.613	0.609	0.594	0.600
`rvd`	-0.076	-0.067	-0.096	-0.119
`jaccard`	0.634	0.621	0.604	0.614

the code is built on Image_Segmentation

Later work

[ ] data augmentation
[x] postprocessing (Just according to some tips on nnUnet,if you want to do some others like CRF,you can make it by yourself on pydensecrf)
[x] 3d segmentation

Some others Related Work on Liver tumor segmentation

3d model for segmentation
3DUNet-Pytorch
MICCAI-LITS2017
RAUNet-tumor-segmentation
LiTS---Liver-Tumor-Segmentation-Challenge
H-DenseUNet
DoDNet
lits
SegWithDistMap

2d model for segmentation
LiTS-Liver-Tumor-Segmentaton-Challenge
ISBI-2020-LITS_Hybrid_Comp_Net
unet-lits-2d-pipeline
DS-SFFNet
u_net_liver

zz10001 / LITS2017-main1

readme