Vahadane Stain Normalization

Jaume, G., Pati, P., Anklin, V., Foncubierta, A., & Gabrani, M. (2021). HistoCartography: A Toolkit for Graph Analytics in Digital Pathology (arXiv:2107.10073). arXiv. http://arxiv.org/abs/2107.10073

Histocartography paper - Vahadane Stain Normalization Paper

---

Vahadane Stain Normalization is one of the most popular preprocessing methods in digital pathology. As stated in the paper, "this normalizer is a solution for stain separation and color normaization, which preserves biological structure information by modelling stain density maps"

What does this container do?

This container will expose a port on your machine/cloud instance which will accept images via an API, and return a .zip file containing the normalized and split images.

More technically, each image passed into the API will be split into 8, separate pieces (feature added soon to dynamically change this). Each image will then be passed in to the Vahadane Stain Normalization algorithm (learn more here) (feature added soon to dynamically change the stain algorithm used). The images will then be zipped and sent in a file titled normalizedAndSplitPictures.zip. Check out main.py for more details

We have 3 images in the images/ directory where you can test out the container. We also have the files on Bacalhau, which you can learn more about here.

Running this app locally

Run the app locally with: uvicorn main:app --reload --host "0.0.0.0"

Building the container image

1. Clone this repository

2. Inside the directory for this repository, run:

   docker build -t vahadane_norm_and_tile .
   

Running the container

Run the app in Docker via:

docker run -d -p 8000:8000 vahadane_norm_and_tile:latest

and it should be running at http://localhost:8000 (or replace localhost with the IP/DNS name of whichever computer is running the container)

Processing data with Postman

1. In Postman, craft a request with the following properties:

   Postman parameter	Value
   Request Type	POST
   Body Type	form-data
   key type (dropdown)	File
   key name (textbox)	files
   key value	(select an image)

2. Press the "Send" button

3. The result should be a .zip file, so when the result is ready (it will look like random letters and numbers in the bottom half of the screen), press "Save Response" on the right side of the screen. Save the zip file.

Processing data with bacalhau

You can learn more about bacalhau in the docs.

the examples images have been pinned to IPFS using Pinata:

QmTr26AJDQMyxEJacL2ph7N71SAgRJaqj4USt4bMiE4Jff
QmX4kSU5W5Lnjjx9ZhBezonMdKQiy3zbJfkSLfUE4cTDMd
QmQr9LBQhmCUzRXkZZeN5nVZriavubLpBffFv5WkBFNUFU

Install bacalhau via:

curl -sL https://get.bacalhau.org/install.sh | bash

To process the first of the example images using bacalhau, run:

bacalhau docker run \                                         
  -v QmTr26AJDQMyxEJacL2ph7N71SAgRJaqj4USt4bMiE4Jff:/inputs/test.png \
  ghcr.io/openlab-apps/vahadane_norm_and_tile:main python main.py /inputs/test.png 

You can download the results by calling bacalhau get on the job id you have received.