2nd Place Submission to ISBI 2020's MoNuSAC Post-Challenge

Hover-Net adapted to the Multi-Task, Transfer-Learning setting

To obtain instance-masks from the trained model on the MoNuSAC test data in the required challenge format:

• Ensure library requirements mentioned in requirements.txt are installed
• Download MoNuSAC test-data & trained model
• Modify data_dir, output-dir, model-path to reflect MoNuSAC test data, desired output path, & downloaded trained model path (.index file) respectively.

Run the below script as:

python test_script.py --data_dir --output_dir --model_path --img_ext --gpu

Getting Started

Install the required libraries before using this code. Please refer to requirements.txt

Dataset

Download the datasets & modify the paths in config_multitask:

• CoNSeP
• Kumar & CPM-17

Ground truth files are in .mat format, refer to the README included with the datasets for further information.

Repository

Files needed to modify multi-task behavior:

• test_script : Test & Eval script
• config_multitask : Config
• train_multitask : Training Script
• hover_multitask : Network & Training protocol

Global Repository Structure

For details on the general repository structure, refer to hover_net-README.md

Citations

MoNuSAC 2020

@article{monusac2020,
author = {Verma, Ruchika; Kumar, Neeraj; Patil, Abhijeet; Kurian, Nikhil; Rane, Swapnil; and Sethi, Amit},
year = {2020},
month = {02},
pages = {},
language = {en},
title = {Multi-organ Nuclei Segmentation and Classification Challenge 2020},
publisher = {Unpublished},
doi = {10.13140/RG.2.2.12290.02244/1},
 url = {http://rgdoi.net/10.13140/RG.2.2.12290.02244/1}
}

HoVer-Net Paper [Linked here]

@article{graham2019hover,
  title={Hover-net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images},
  author={Graham, Simon and Vu, Quoc Dang and Raza, Shan E Ahmed and Azam, Ayesha and Tsang, Yee Wah and Kwak, Jin Tae and Rajpoot, Nasir},
  journal={Medical Image Analysis},
  pages={101563},
  year={2019},
  publisher={Elsevier}
}