KiTS19——2019 Kidney Tumor Segmentation Challenge

This is an example of the CT images Kidney Tumor Segmentation

Prerequisities

The following dependencies are needed:

• python == 3.5.5
• numpy >= 1.11.1
• SimpleITK >= 1.0.1
• opencv-python >= 3.3.0
• tensorflow-gpu == 1.8.0
• pandas >=0.20.1
• scikit-learn >= 0.17.1
• json >=2.0.9

How to Use

1、Preprocess

• analyze the ct image,and get the slice thickness and window width and position:run the dataAnaly.py

1.1 Preprocess Kidney

• keep kidney region into fixed size(512,512,64) for Corse Kidney Segmentation:run the corsedata2dprepare.py
• generate patch(128,128,64) kidney image and mask for Corse Kidney Segmentation:run the corsedata3dprepare.py
• keep Kidney region range image for fine Kidney Segmentation:run the finedata2dprepare.py
• generate patch(128,128,64) kidney image and mask for fine Kidney Segmentation:run the finedata3dprepare.py
• save patch image and mask into csv file: run the utils.py,like file trainSegmentation.csv
• split trainSegmentation.csv into training set and test set:run subset.py

1.2 Preprocess Kidney Tumor

• generate tumor image and mask for 2d Kidney Tumor Segmentation:run the tumordata2dprepare.py
• generate tumor image and mask for 3d Kidney Tumor Segmentation:run the finedata3dprepare.py
• save tumor image and mask path into csv file: run the utils.py,like file traintumorSegmentation.csv
• split traintumorSegmentation.csv into training set and test set

2、Kidney Segmentation

• the VNet model

2.1 Corse Kidney Segmentation

• Corse Kidney Segmentation training:run the train_vnet3d_kidney_corse.py
• Corse Kidney Segmentation inference:run the inference_vnet3d_kidney_corse.py
• this step get Corse Kidney range,can find the start and end pos in the kidneyrang.txt

2.2 Fine Kidney Segmentation

• Fine Kidney Segmentation training:run the train_vnet3d_kidney_fine.py
• Fine Kidney Segmentation inference:run the inference_vnet3d_kidney_fine.py
• this step following the 2.1 result,get fine Kidney result

2.3 Fine Kidney Segmentation

• remove Kidney Segmentation small object:run the segresultprocess.py removekidneysmallobj function

3、Kidney Tumor Segmentation

• the VNet2d model

3.1 2d Kidney Tumor Segmentation

• 2d Kidney Tumor Segmentation training:run the train_vnet2d_tumor.py

• 2d Kidney Tumor Segmentation inference:run the inference_vnet2d_tumor.py

• this step get 2d slice tumor result

• the VNet3d model

3.2 3d Kidney Tumor Segmentation

• 3d Kidney Tumor Segmentation training:run the train_vnet3d_tumor.py
• 3d Kidney Tumor Segmentation inference:run the inference_vnet3d_tumor.py
• this step get 3d tumor result

3.3 Kidney Tumor Result Process

• remove Kidney Tumor Segmentation small object:run the segresultprocess.py remove2d3dtumorsmallobj function
• calculate overlap between 2d tumor and 3d tumor reslut.
• save the region of 2d tumor and 3d tumor reslut that connect overlap region.
• save the region of 2d tumor and 3d tumor within Kidney result.
• merge the above two result and get final tumor result
• all above step can find in the segresultprocess.py

Result

1、Kidney Segmentation

• the train loss

• 200-209case dice value and result

2、Kidney Tumor Segmentation

• the 2dtrain loss

• the 3dtrain loss

3、Lead Board

Contact

• https://github.com/junqiangchen
• email: [email protected]，[email protected]
• Contact: junqiangChen，junMa(马骏)
• WeChat Number: 1207173174
• WeChat Public number: 最新医学影像技术