/SegTHOR2019

SegTHOR2019 SegTHOR Challenge: Segmentation of THoracic Organs at Risk in CT Images

Primary LanguageJupyter Notebook

segthor for keras

比赛网址 for more information: https://competitions.codalab.org/competitions/21012

Tumor Segmentation Example

label

标签 value nickname
食管 1 red
心脏 2 pink
气管 3 yellow
动脉 4 blue

最后一次的提交结果:

时间 食管 心脏 气管 主动脉
3月17日 0.8452(第五名) 0.9328(第十八) 0.9039(第十四) 0.9115(第十六)
3月22日 0.8513(第七名) 0.9457(第九名) 0.9083(第十五) 0.9175(第二十四)

--
目前有优于以上的成绩的模型。

各个任务的相关实验数据 :

食管: RED
心脏: PINK
气管: YELLOW
主动脉: BLUE

Data understanding

Example Pic

Sample

example

Training & Testing Data

The whole SegTHOR dataset (60 patients and 11084 slices) has been randomly split into:

  • a training set: 40 patients, 7390 slices
  • a testing set: 20 patients, 3694 slices
graph LR;
A[train set 40]-->B[for val 8];
A-->C[for train 32];
Loading

Train

Seg and cls mdoel

Inference https://github.com/qubvel/segmentation_models

Avaliable backbones:

Backbone model Name Weights
VGG16 vgg16 imagenet
VGG19 vgg19 imagenet
ResNet18 resnet18 imagenet
ResNet34 resnet34 imagenet
ResNet50 resnet50 imagenet
imagenet11k-places365ch
ResNet101 resnet101 imagenet
ResNet152 resnet152 imagenet
imagenet11k
ResNeXt50 resnext50 imagenet
ResNeXt101 resnext101 imagenet
DenseNet121 densenet121 imagenet
DenseNet169 densenet169 imagenet
DenseNet201 densenet201 imagenet
Inception V3 inceptionv3 imagenet
Inception ResNet V2 inceptionresnetv2 imagenet

Data processing

Data auto crop

graph LR;
F[3D cube]-->E(确定图像中心);
E-->V(3D crop)
V-.flatten.->G[单张slice]
Loading

Data normalization for CT

graph LR;
F[train CT image]-->E[HU mean for ev Pat];
L[train mask]-->E[ww and wl];
E-->G(train for norm CT image)
H[test  CT image]-->Q[ww and wl];
O[test  mask predict by older model]-->Q[ww and wl];
Q-->I(test for norm CT image)
Loading

效果图:

Sample Sample

example