NucleiSegHE

H&E ROI-Level and WSI-Level Nuclei Segmentation with HoVer-Net - pretrained model and demo ROIs/WSIs.

Environment Configurations

a. Prepare docker image

• Build from Dockerfile

$ docker build -t nucleiseghe:chen .

• Or pull from Docker Hub

$ docker pull pingjunchen/nucleiseghe:chen
$ docker tag pingjunchen/nucleiseghe:chen nucleiseghe:chen

b. Setup docker container

• Start docker container (specify CODE_ROOT & DATA_ROOT)

$ docker run -it --rm --user $(id -u):$(id -g) \
  -v ${CODE_ROOT}:/App/NucleiSegHE \
  -v ${DATA_ROOT}:/Data \
  --shm-size=224G --gpus '"device=0,1"' --cpuset-cpus=0-15 \
  --name nucleiseghe_chen nucleiseghe:chen

• For example:

$ docker run -it --rm  --user $(id -u):$(id -g) \
  -v /rsrch1/ip/pchen6/Codes/CHEN/NucleiSegHE:/App/NucleiSegHE \
  -v /rsrch1/ip/pchen6/NucleiSegData:/Data \
  --shm-size=224G --gpus '"device=0,1"' --cpuset-cpus=0-15 \
  --name nucleiseghe_chen nucleiseghe:chen

ROI-Level Nuclei Seg (support png)

Inside the docker container, enter /App/NucleiSegHE

# Nuclei Segmentation
$ python 01_roi_seg_nuclei.py --dataset LungNYU
# Nuclei Overlay
$ python 02_roi_nuclei_overlay.py --dataset LungNYU

WSI-Level Nuclei Seg (support svs/tiff)

Inside the docker container, enter /App/NucleiSegHE

# Split WSI into smaller blocks (5000 x 5000)
$ python 00_wsi_split_blocks.py --dataset CLL
# Block-wise WSI nuclei segmentation and merging
$ python 01_wsi_seg_nuclei.py --dataset CLL
# Nuclei overlay to the entire WSI
$ python 02_wsi_nuclei_overlay.py --dataset CLL

Acknowledgements

This repo is adapted from following codes

• vqdang/hover_net
• simongraham/hovernet_inference

Please consider citing the following two papers if this repo was used for nuclei segmentation in your research

• Hover-Net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images
• PanNuke: an open pan-cancer histology dataset for nuclei instance segmentation and classification