Whole slide image representation in bone marrow cytology

This a Github Repo hosting custom codes for the paper "Whole slide image representation in bone marrow cytology".

System requirements

Hardware Requirements

For optimal performance, we recommend a computer with the following specs:

RAM: 16+ GB
CPU: 2+ cores, 2.2+ GHz/core
GPU: 16+ GB

The runtimes below are generated using a computer with the recommended specs:

RAM: 16 GB
CPU: 2 Intel(R) Xeon(R) CPU @ 2.20GHz
GPU: 1 Tesla V100-SXM2-16GB, CUDA Version: 10.1

Software Requirements

The package development version is tested on Linux operating system (Ubuntu 18.04.5 LTS).

Python Dependencies:

python = ">=3.9,<3.11"
faiss-cpu = ">=1.7.2"
torch = ">=1.10.2"
numpy = ">=1.18"
record-keeper = ">=0.9.31"
tensorboard = ">=2.8.0"
plotly = ">=5.6.0"
scikit-learn = ">=1.0.2"
numba = ">=0.55.1"
scipy = "^1.8.0"
umap-learn = ">=0.5.2"
pytorch-metric-learning = ">=1.2.0"
tqdm = "^4.63.0"
kaleido = "^0.2.1"

It takes about 5-10 mins.

Demo

The data that support the findings of this study are available on reasonable request from the corresponding author, pending local REB and privacy office approval. The data are not publicly available because they contain information that could compromise research participant privacy/consent.

You need first to contact the corresponding author to get the zipfile data.zip used in for this demo

Instructions for use

Scripts

Modeling process: modeling.py
Refined HCT alone benchmark: hct_clf.py
Metrics Figure come from clf_result.py
Attention weight plot plot_att_weight.py
Copy materials for interactive plot copy_att.py
Embedding Figure pick_embedding.py
Cell type masking im_score.py
A preview of Cell bag plot_cell_bag.py