Skip to content

1st place solution to the Breast Cancer Classification Task of HeLP Challenge 2019.

Notifications You must be signed in to change notification settings

ExcelsiorCJH/HeLP2019_Breast_Cancer_1st_solution

 
 

Repository files navigation

HeLP Challenge 2019 Breast Cancer 1st place solution

This repository is 1st place solution to the Breast Cancer Classification Task of HeLP Challenge 2019.
task_description

Model

model_description

Stage 1

  • Preprocessing: ROI extraction, Rescale, Vahadane Stain Normalization
  • Pixel-wise Segmentation: Feature Pyramid Network(FPN)

Stage 2

  • Feature extraction from probability heatmap
  • Prediction final probability and major axis based on features

And also, please click this link to see the detailed model description.

Dependencies

  • keras
  • segmentation_models
  • openslide
  • staintools
  • numpy
  • pandas
  • sklearn
  • skimage

Usage

Dataset

data
  └── train
     ├── level4
     │  ├── Image
     │  │  ├── slide_001.png
     │  │  ├── ...
     │  │  └── slide_#.png
     │  └── Mask
     │     ├── mask_001.png
     │	   ├── ...
     │	   └── mask_#.png
     └── label.csv
            
========= After training, the directories are created as below. =========

  ├── volume
  │  ├── dataset
  │  │  └── level4 
  │  │     ├── img
  │  │	   │  ├── slide001_patch001.png
  │  │ 	   │  ├── ...
  │  │     │  └── slide#_patch#.png
  │  │	   └── mask
  │  │	      ├── mask001_patch001.png
  │  │        ├── ...
  │  │        └── mask#_patch#.png
  │  └── model
  │       └── fpn_weights.h5
  └── heatmap
      ...

Train

Run the train.py.

$ python train.py

Inference

Run the inference.sh.

$ sh inference.sh

Authors

About

1st place solution to the Breast Cancer Classification Task of HeLP Challenge 2019.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.9%
  • Shell 0.1%