🤖🔬 PathML: Tools for computational pathology
⭐ PathML objective is to lower the barrier to entry to digital pathology
Imaging datasets in cancer research are growing exponentially in both quantity and information density. These massive datasets may enable derivation of insights for cancer research and clinical care, but only if researchers are equipped with the tools to leverage advanced computational analysis approaches such as machine learning and artificial intelligence. In this work, we highlight three themes to guide development of such computational tools: scalability, standardization, and ease of use. We then apply these principles to develop PathML, a general-purpose research toolkit for computational pathology. We describe the design of the PathML framework and demonstrate applications in diverse use cases.
🚀 The fastest way to get started?
docker pull pathml/pathml && docker run -it -p 8888:8888 pathml/pathml
done, what analyses can I write now? 👉
This AI will: - 🤖 write digital pathology analyses for you - 🔬 walk you through the code, step-by-step - 🎓 be your teacher, as you embark on your digital pathology journey ❤️ More usage examples [here](./ai-digital-pathology-assistant-v3). |
📖 Official PathML Documentation
View the official PathML Documentation on readthedocs
🔥 Examples! Examples! Examples!
↴ Jump to the gallery of examples below
There are several ways to install PathML
:
pip install
from PyPI (recommended for users)Options (1), (2), and (4) require that you first install all external dependencies:
We recommend using conda for environment management. Download Miniconda here
Create conda environment, this step is common to all platforms (Linux, Mac, Windows):
conda create --name pathml python=3.8
conda activate pathml
Install external dependencies (for Linux) with Apt:
sudo apt-get install openslide-tools g++ gcc libblas-dev liblapack-dev
Install external dependencies (for MacOS) with Brew:
brew install openslide
Install external dependencies (for Windows) with vcpkg:
vcpkg install openslide
Install OpenJDK 8, this step is common to all platforms (Linux, Mac, Windows):
conda install openjdk==8.0.152
Optionally install CUDA (instructions here)
Install PathML
from PyPI:
pip install pathml
Clone repo:
git clone https://github.com/Dana-Farber-AIOS/pathml.git
cd pathml
Create conda environment:
conda env create -f environment.yml
conda activate pathml
Optionally install CUDA (instructions here)
Install PathML
from source:
pip install -e .
First, download or build the PathML Docker container:
Step 1: download PathML container from Docker Hub
docker pull pathml/pathml:latest
Optionally specify a tag for a particular version, e.g. docker pull pathml/pathml:2.0.2
. To view possible tags,
please refer to the PathML DockerHub page.
Alternative Step 1 if you have custom hardware: build docker container from source
git clone https://github.com/Dana-Farber-AIOS/pathml.git
cd pathml
docker build -t pathml/pathml .
Step 2: Then connect to the container:
docker run -it -p 8888:8888 pathml/pathml
The above command runs the container, which is configured to spin up a jupyter lab session and expose it on port 8888.
The terminal should display a URL to the jupyter lab session starting with http://127.0.0.1:8888/lab?token=<.....>
.
Navigate to that page and you should connect to the jupyter lab session running on the container with the pathml
environment fully configured. If a password is requested, copy the string of characters following the token=
in the
url.
Note that the docker container requires extra configurations to use with GPU.
Note that these instructions assume that there are no other processes using port 8888.
Please refer to the Docker run
documentation for further instructions
on accessing the container, e.g. for mounting volumes to access files on a local machine from within the container.
To get PathML running in a Colab environment:
import os
!pip install openslide-python
!apt-get install openslide-tools
!apt-get install openjdk-8-jdk-headless -qq > /dev/null
os.environ["JAVA_HOME"] = "/usr/lib/jvm/java-8-openjdk-amd64"
!update-alternatives --set java /usr/lib/jvm/java-8-openjdk-amd64/jre/bin/java
!java -version
!pip install pathml
PathML Tutorials we published in Google Colab
Thanks to all of our open-source collaborators for helping maintain these installation instructions!
Please open an issue for any bugs or other problems during installation process.
To use GPU acceleration for model training or other tasks, you must install CUDA. This guide should work, but for the most up-to-date instructions, refer to the official PyTorch installation instructions.
Check the version of CUDA:
nvidia-smi
Install correct version of cudatoolkit
:
# update this command with your CUDA version number
conda install cudatoolkit=11.0
After installing PyTorch, optionally verify successful PyTorch installation with CUDA support:
python -c "import torch; print(torch.cuda.is_available())"
Jupyter notebooks are a convenient way to work interactively. To use PathML
in Jupyter notebooks:
PathML relies on Java to enable support for reading a wide range of file formats.
Before using PathML
in Jupyter, you may need to manually set the JAVA_HOME
environment variable
specifying the path to Java. To do so:
echo $JAVA_HOME
in the terminal in your pathml conda environment (outside of Jupyter)JAVA_HOME
environment variable in Jupyter:
import os
os.environ["JAVA_HOME"] = "/opt/conda/envs/pathml" # change path as needed
conda activate pathml
conda install ipykernel
python -m ipykernel install --user --name=pathml
This makes the pathml environment available as a kernel in jupyter lab or notebook.
Now that you are all set with PathML
installation, let's get started with some analyses you can easily replicate:
1. [Load over 160+ different types of pathology images using PathML](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/loading_images_vignette.ipynb) 2. [H&E Stain Deconvolution and Color Normalization](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/stain_normalization.ipynb) 3. [Brightfield imaging pipeline: load an image, preprocess it on a local cluster, and get it read for machine learning analyses in PyTorch](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/workflow_HE_vignette.ipynb) 4. [Multiparametric Imaging: Quickstart & single-cell quantification](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/multiplex_if.ipynb) 5. [Multiparametric Imaging: CODEX & nuclei quantization](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/codex.ipynb) 6. [Train HoVer-Net model to perform nucleus detection and classification, using data from PanNuke dataset](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/train_hovernet.ipynb) 7. [Gallery of PathML preprocessing and transformations](https://github.com/Dana-Farber-AIOS/pathml/blob/master/examples/pathml_gallery.ipynb) |
If you use PathML
please cite:
So far, PathML was referenced in 20+ manuscripts:
This is where in the world our most enthusiastic supporters are located:
|
and this is where they work:
|
Source: https://ossinsight.io/analyze/Dana-Farber-AIOS/pathml#people
PathML
is an open source project. Consider contributing to benefit the entire community!
There are many ways to contribute to PathML
, including:
PathML
with colleagues, students, etc.See contributing for more details.
The GNU GPL v2 version of PathML is made available via Open Source licensing. The user is free to use, modify, and distribute under the terms of the GNU General Public License version 2.
Commercial license options are available also.
Questions? Comments? Suggestions? Get in touch!