Skip to content
This repository has been archived by the owner on Jan 18, 2022. It is now read-only.

Latest commit

 

History

History
28 lines (17 loc) · 2.13 KB

README.md

File metadata and controls

28 lines (17 loc) · 2.13 KB

Update 20220118

This repo is archived since there is no updates, and I'm working on it anymore. For any question about the challenge, ask the organizers for help.

PSPNet for MICCAI Automatic Prostate Gleason Grading Challenge 2019

This is the code for MICCAI Automatic Prostate Gleason Grading Challenge 2019. Check here and here, we took the 1st place of task 1: pixel-level Gleason grade prediction and task 2: core-level Gleason score prediction (leaderboard).

Task 1 is regarded as a segmentation task, and we use PSPNet for this. And for task 2, we do not train a different network, but just produce the prediction from the prediction of task 1 according to the Gleason grading system.

The train script is based on reference script from torchvision 0.4.0 with minor modification. So, you need to install the latest PyTorch and torchvision >= 0.4.0. Check requirements.txt for all packages you need.

This repo use GluonCV-Torch, thanks for Hang Zhang's outstanding work!

Preprocessing

Each image is annotated in detail by several expert pathologists. So how to use this annotations is important. We use STAPLE to create final annotations used in training. Check the preprocessing.py script for detail.

preprocessing

Training

PSPNet

To run the training, simply run python train.py, check python train_gleason.py --help for available args.

Inference

To run the inference, simply run python inference.py, check python inference.py --help for available args.

Note

I don't quite understand task2, and got it wrong when I participated this challenge. I would sincerely advise you read this paper, which is written by the organizers and submitted to JBHI, for more detail about this challenge. I would not update any codes in this repository anymore.