|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# Checkpoint for Multistain Pretraining for Slide Representation Learning in Pathology (ECCV'24) |
|
|
|
|
|
Welcome to the official HuggingFace repository of the ECCV 2024 paper, "Multistain Pretraining for Slide Representation Learning in Pathology". |
|
|
This project was developed at the [Mahmood Lab](https://faisal.ai/) at Harvard Medical School and Brigham and Women's Hospital. |
|
|
|
|
|
## Model loging |
|
|
|
|
|
```python |
|
|
from huggingface_hub import login |
|
|
login() |
|
|
``` |
|
|
|
|
|
You can refer [HuggingFace](https://huggingface.co/docs/huggingface_hub/en/quick-start#login-command) documentation for more details. |
|
|
|
|
|
## Preprocessing: tissue segmentation, patching, and patch feature extraction |
|
|
|
|
|
We are extracting [CONCH](https://github.com/mahmoodlab/CONCH) features at 10x magnification on 256x256-pixel patches. Please refer to Madeleine public implementation to extract patch embeddings from a WSI. |
|
|
|
|
|
## Extracting MADELEINE slide encoding |
|
|
|
|
|
You can obtain and run MADELEINE slide encoding (trained on Acrobat breast samples at 10x magnification) using: |
|
|
|
|
|
```python |
|
|
from core.models.factory import create_model_from_pretrained |
|
|
|
|
|
model, precision = create_model_from_pretrained('./models/') |
|
|
|
|
|
feats = load_h5('your_path_to_conch_patch_embeddings/XXX.h5') |
|
|
|
|
|
with torch.no_grad(): |
|
|
with torch.amp.autocast(device_type="cuda", dtype=precision): |
|
|
wsi_embed = model.encode_he(feats=feats, device='cuda') |
|
|
``` |
|
|
|
