Skip to content

AICUP 2022 Identification of Spread through Air Spaces (STAS) in Pathology Images of Lung Adenocarcinoma (II): Using Image Segmentation Strategies to Recognize STAS Contour

Notifications You must be signed in to change notification settings

JW-Shen/AICUP-2022-Identification-of-Spread-through-Air-Spaces-STAS-in-Pathology-Images-of-Lung-Adenocarci

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

32 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI CUP 2022 Identification of Spread through Air Spaces (STAS) in Pathology Images of Lung Adenocarcinoma (II): Using Image Segmentation Strategies to Recognize STAS Contour

針對肺腺癌 H&E 染色數位病理全切片影像,本競賽提供在腫瘤外的感興趣區域 (region of interest, ROI) 以方框及不規則形狀像素層級之 STAS 標註資訊,運用影像分割作法於切割STAS輪廓。

  • TEAM_1277
  • Private leaderboard:0.90546 / (Rank 13 / 307)

Preprocessing (for cross validation)

  • 將資料分群讓不同特性的資料均勻分布各個 cross validation 的資料集 , 以增加 ensemble 後的模型穩定度
    • way1 : 肉眼觀察分群 (資料集照一定順序分佈)
    • way2 : kmeans cluster
      • step1 : Elbow Method 決定分4群最佳 (註: 用 keras inceptionV3 抽特徵)
      • step2 : 4 群均分到不同 training dataset Untitled (2) Untitled (3)

Model Architecture (voting)

  • 使用 segmentation_models_pytorch 所提供的 API 建立以下五個模型
    Architectures Encoders
    UnetPlusPlus Efficientnet b7
    UnetPlusPlus Efficientnet b5
    UnetPlusPlus SE-ResNeXt50-32x4d
    DeepLabV3Plus Efficientnet b5
    Linknet Efficientnet b5
  • 最後將五個模型 output 做平均 (Average Voting)

Postprocessing (blur + findcontours & fillpoly)

image

repo 解說

前處理程式碼:Preprocessing.ipynb

訓練程式碼:Model.ipynb

辨識程式碼:Test.ipynb

模型檔案:https://drive.google.com/drive/folders/1Sq682KheFmneDXpLD5Y7cYXpxpHj-gup?usp=sharing

執行環境:TWCC

預測結果輸出:STAS.zip

執行環境:Pytorch 1.11.0

檔案目錄結構

├── Preprocessing.ipynb               
├── Model.ipynb
├── Test.ipynb
├── Data
│   ├── SEG_Train_Datasets
│   │   ├── Train_Images
│   │   ├── Train_Annotations
│   │   ├── Train_Masks
│   │   ├── Fold1_Images
│   │   ├── Fold1_Masks
│   │   ├── Fold2_Images
│   │   ├── Fold2_Masks
│   │   ├── Fold3_Images
│   │   ├── Fold3_Masks
│   │   ├── Fold4_Images
│   │   ├── Fold4_Masks
│   │   ├── Test_Images
│   │   └──  Test_Masks
│   ├── Public_Image
│   └── Image   
├── model_weight                      
│   ├── best_model_1.pth
│   ├── best_model_2.pth               
│   ├── best_model_3.pth         
│   ├── best_model_4.pth               
│   └── best_model_5.pth
├── fig                      
│   ├── acc_1.png
│   ├── acc_2.png              
│   ├── acc_3.png         
│   ├── acc_4.png              
│   ├── acc_5.png
│   ├── loss_1.png
│   ├── loss_2.png              
│   ├── loss_3.png         
│   ├── loss_4.png              
│   └── loss_5.png
├── output                      
│   ├── Image           
│   └── Image_Postprocess

About

AICUP 2022 Identification of Spread through Air Spaces (STAS) in Pathology Images of Lung Adenocarcinoma (II): Using Image Segmentation Strategies to Recognize STAS Contour

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published