Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling, MedRxiv, July 2023.
Filiot, A., Ghermi, R., Olivier, A., Jacob, P., Fidon, L., Kain, A. M., Saillard, C., & Schiratti, J.-B. (2023). Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling. MedRxiv.
@article{Filiot2023scalingwithMIM,
author = {Alexandre Filiot and Ridouane Ghermi and Antoine Olivier and Paul Jacob and Lucas Fidon and Alice Mac Kain and Charlie Saillard and Jean-Baptiste Schiratti},
title = {Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling},
elocation-id = {2023.07.21.23292757},
year = {2023},
doi = {10.1101/2023.07.21.23292757},
publisher = {Cold Spring Harbor Laboratory Press},
url = {https://www.medrxiv.org/content/early/2023/07/26/2023.07.21.23292757v2},
eprint = {https://www.medrxiv.org/content/early/2023/07/26/2023.07.21.23292757v2.full.pdf},
journal = {medRxiv}
}
We released our Phikon model on Hugging Face. Check out our community blog post ! We also provide a Colab notebook to perform weakly-supervised learning on Camelyon16 and fine-tuning with LoRA on NCT-CRC-HE using Phikon.
Here is a code snippet to perform feature extraction using Phikon.
from PIL import Image
import torch
from transformers import AutoImageProcessor, ViTModel
# load an image
image = Image.open("assets/example.tif")
# load phikon
image_processor = AutoImageProcessor.from_pretrained("owkin/phikon")
model = ViTModel.from_pretrained("owkin/phikon", add_pooling_layer=False)
# process the image
inputs = image_processor(image, return_tensors="pt")
# get the features
with torch.no_grad():
outputs = model(**inputs)
features = outputs.last_hidden_state[:, 0, :] # (1, 768) shape
Official PyTorch Implementation and pre-trained models for Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling
. This minimalist repository aims to:
- Publicly release the weights of our Vision Transformer Base (ViT-B) model Phikon pre-trained with iBOT on 40M pan-cancer histology tiles from TCGA. Phikon achieves state-of-the-art performance on a large variety of downstream tasks compared to other SSL frameworks available in the literature.
From 09.01.2023 to 10.30.2023, this repository stated using the student
, please use the teacher
backbone instead.
# feature extraction snippet with `rl_benchmarks` repository
from PIL import Image
from rl_benchmarks.models import iBOTViT
# instantiate iBOT ViT-B Pancancer model, aka Phikon
# /!\ please use the "teacher" encoder which produces better results !
weights_path = "/<your_root_dir>/weights/ibot_vit_base_pancan.pth">
ibot_base_pancancer = iBOTViT(architecture="vit_base_pancan", encoder="teacher", weights_path=weights_path)
# load an image and transform it into a normalized tensor
image = Image.open("assets/example.tif") # (224, 224, 3), uint8
tensor = ibot_base_pancancer.transform(image) # (3, 224, 224), torch.float32
batch = tensor.unsqueeze(0) # (1, 3, 224, 224), torch.float32
# compute the 768-d features
features = ibot_base_pancancer(batch).detach().cpu().numpy()
assert features.shape == (1, 768)
- Publicly release the histology features of our ViT-based iBOT models (
iBOT[ViT-S]COAD
,iBOT[ViT-B]COAD
,iBOT[ViT-B]PanCancer
,iBOT[ViT-L]COAD
) for i) 11 TCGA cohorts and Camelyon16 slides datasets; and ii) NCT-CRC and Camelyon17-Wilds patches datasets. - Reproduce the results from our publication, including: features extraction and clinical data processing, cross-validation experiments, results generation.
Read full abstract from MedRxiv.
Computational pathology is revolutionizing the field of pathology by integrating advanced computer vision and machine learning technologies into diagnostic workflows. Recently, Self-Supervised Learning (SSL) has emerged as a promising solution to learn representations from histology patches, leveraging large volumes of unannotated whole slide images whole slide images (WSI). In particular, Masked Image Modeling (MIM) showed remarkable results and robustness over purely contrastive learning methods. In this work, we explore the application of MIM to histology using iBOT, a self-supervised transformer-based framework. Through a wide range of downstream tasks over seven cancer indications, we provide recommendations on the pre-training of large models for histology data using MIM. First, we demonstrate that in-domain pre-training with iBOT outperforms both ImageNet pre-training and a model pre-trained with a purely contrastive learning objective, MoCo V2. Second, we show that Vision Transformers (ViT), when scaled appropriately, have the capability to learn pan-cancer representations that benefit a large variety of downstream tasks. Finally, our iBOT ViT-Base model, pre-trained on more than 40 million histology images from 16 different cancer types, achieves state-of-the-art performance in most weakly-supervised WSI classification tasks compared to other SSL frameworks. Our code, models and features are publicly available at https://github.com/owkin/HistoSSLscaling.You can download the data necessary to use the present code and reproduce our results here:
- raw data: Google Drive
- preprocessed data: Google Drive
- weights: Google Drive
Please create weights
, raw
and preprocessed
folders containing the content of the different downloads. This step may take time depending on your wifi bandwidth (folder takes 1.2 To). You can use rclone to download the folder from a remote machine (preferred in a tmux
session).
The bucket contains three main folders: a weights
, raw
and preprocessed
folders. The weights
folder contains weights for iBOT[ViT-B]PanCancer
(our best ViT-B iBOT model). Other models from the literature can be retrieved from the corresponding Github repositories:
- CTransPath: https://github.com/Xiyue-Wang/TransPath
- HIPT: https://github.com/mahmoodlab/HIPT
- Dino[ViT-S]BRCA: https://github.com/Richarizardd/Self-Supervised-ViT-Path
weights/
βββ ibot_vit_base_pancan.pth # Ours
The raw
folder contains two subfolders for slide-level and tile-level downstream task.
- Slide-level: each cohort contains 2 folders,
clinical
andslides
. We provide clinical data but not raw slides. No modification was performed on the folders architectures and files names of raw slides and patches compared to the original source (i.e. TCGA, Camelyon16, NCT-CRC and Camelyon17-WILDS). - Tile-level: each cohort contains 2 folders,
clinical
andpatches
. We only provide clinical data (i.e. labels), not patches datasets.
Warning
We don't provide raw slides or patches (slides
, patches
folders are empty).
You can download raw slides or patches here:
- PAIP: https://www.wisepaip.org/paip/guide/dataset
- TCGA: https://portal.gdc.cancer.gov/
- Camelyon16: https://gigadb.org/dataset/100439
- NCT-CRC: https://zenodo.org/record/1214456
- Camelyon17-WILDS: https://github.com/p-lambda/wilds/blob/main/wilds/download_datasets.py
Once you downloaded the data, please follow the same folders architecture as indicated below (without applying modifications on folders and files names compared to original download).
raw/
βββ slides_classification # slides classification tasks
===============================================================================
βΒ Β βββ CAMELYON16_FULL # cohort
βΒ Β βΒ Β βββ clinical # clinical data (for labels)
βΒ Β βΒ Β βΒ Β βββ test_clinical_data.csv
βΒ Β β βΒ Β βββ train_clinical_data.csv
βΒ Β βΒ Β βββ slides # raw slides (not provided)
βΒ Β βΒ Β βββ Normal_001.tif
βΒ Β βΒ Β Β βββ Normal_002.tif...
βΒ Β βββ TCGA
βΒ Β βββ tcga_statistics.pk # For each cohort and label, list (n_patients, n_slides, labels_distribution)
βΒ Β Β βββ clinical # for TCGA, clinical data is divided into subfolders
βΒ Β Β βΒ Β βββ hrd
βΒ Β Β βΒ Β βΒ Β Β βββ hrd_labels_tcga_brca.csv
βΒ Β Β β βΒ βββ hrd_labels_tcga_ov.csv
βΒ Β Β βΒ Β βββ msi
βΒ Β Β βΒ Β βΒ Β βββ msi_labels_tcga_coad.csv
βΒ Β Β βΒ Β βΒ Β βββ msi_labels_tcga_read.csv...
βΒ Β Β βΒ Β βββ subtypes
βΒ Β Β βΒ Β βΒ Β βββ brca_tcga_pan_can_atlas_2018_clinical_data.tsv.gz
βΒ Β Β βΒ Β βΒ Β βββ coad_tcga_pan_can_atlas_2018_clinical_data.tsv.gz...
β βΒ Β βββ survival
βΒ Β Β βΒ Β Β Β βββ survival_labels_tcga_brca.csv
βΒ Β Β βΒ Β Β Β βββ survival_labels_tcga_coad.csv...
β βββ slides
βΒ Β Β βββ parafine
βΒ Β Β Β Β βββ TCGA_BRCA
βΒ Β Β Β βΒ βββ 03627311-e413-4218-b836-177abdfc3911
βΒ Β Β Β β βΒ Β βββ TCGA-XF-AAN7-01Z-00-DX1.B8EDF045-604C-48CB-8E54-A60564CAE2AD.svs
...
βββ tiles_classification # tiles classification tasks
===============================================================================
Β Β βββ CAMELYON17-WILDS_FULL # cohort
Β Β βΒ Β βββ clinical # clinical data (for labels)
βΒ Β β Β Β βββ metadata.csv
Β Β βΒ Β βββ patches # patches (not provided)
Β Β β Β Β βββ patient_004_node_4...
β βΒ Β βββ patch_patient_004_node_4_x_10016_y_16704.png...
βββ NCT-CRC_FULL
βββ labels # here the labels are set using the folders architecture
βΒ Β βββ dict_labels.pkl
βββ patches
βββ NCT-CRC-VAL-HE-7K
β βββ ADI...
β β βββ ADI-TCGA-AAICEQFN.tif...
βββ NCT-CRC-HE-100K-NONORM
βββ ADI...
β βββ ADI-AAAFLCLY.tif...
The preprocessed
folder contains two subfolders for slide-level and tile-level downstream tasks.
- Slide-level: for each feature extractor and dataset, we provide coordinates and features. Coordinates are provided as (N_tiles_slide, 3) numpy arrays where the 3 first columns rows correspond to
(tile_level, x_coordinate, y_coordinate)
. Features are provided as (N_tiles_slide, 3+d) numpy arrays, the d last columns being the model's features (3 first are the previous coordinates). Coordinates are meant to extract the same tiles as done in our publication but are not needed for downstream experiments (only features are needed). Note that coordinates are divided intocoords_224
,coords_256
andcoords_4096
, corresponding to 224 x 224 tiles (iBOT, CTransPath and ResNet models), 256 x 256 (Dino models) and 4096 x 4096 (HIPT) tiles, respectively.
Note
We provide all matter tiles for each slide. All tiles were extracted at 0.5 micrometers / pixel (20x magnification) except for CTransPath (mpp = 1.0 following the authors recommendation).
Warning
The tile_level
is computed with openslide.deepzoom.DeepZoomGenerator
through the following schematic syntax:
from openslide import open_slide
from openslide.deepzoom import DeepZoomGenerator
slide = open_slide("<slide_path>")
dzg = DeepZoomGenerator(slide, tile_size=224, overlap=0)
tile = dzg.get_tile(level=17, address=(8, 10))
# this corresponds to coordinates (17, 8, 10) in the coordinates we provide for the given slide
- Tile-level: for each feature extractor and dataset, we provide patches ids and features. Features are (N_patches_dataset, d) numpy arrays and ids take the form of (N_patches_dataset, 1) string numpy array.
Here is a description of the different features and coordinates we provide in the preprocessed
folder.
preprocessed/ # preprocessed data (coords, features)
===============================================================================
βββ slides_classification # slides classification tasks
βΒ Β βββ coords
βΒ Β βΒ Β βββ coords_224 # coordinates for 224 x 224 tiles
βΒ Β βΒ Β βΒ Β βββ CAMELYON16_FULL # cohort
βΒ Β βΒ Β βΒ Β βΒ Β βββ Normal_001.tif # slide_id
βΒ Β βΒ Β βΒ Β βΒ Β Β Β βββ coords.npy # coordinates array (N_tiles_slide, 3)
...
βΒ Β βΒ Β βΒ Β βββ TCGA
βΒ Β βΒ Β βΒ Β β βββ TCGA_BRCA
βΒ Β βΒ Β βΒ Β β βΒ Β βββ TCGA-3C-AALI-01Z-00-DX1.F6E9A5DF-D8FB-45CF-B4BD-C6B76294C291.svs
βΒ Β β βΒ Β βΒ Β βΒ Β Β Β βββ coords.npy
...
βΒ Β βΒ Β βββ coords_256 # coordinates for 256 x 256 tiles
βΒ Β βΒ Β βββ coords_4096 # coordinates for 4096 x 4096 tiles
...
βΒ Β βββ features # features
βΒ Β Β Β βββ iBOTViTBasePANCAN # feature extractor
βΒ Β βΒ Β βββ CAMELYON16_FULL # cohort
βΒ Β βΒ Β βΒ Β βββ Normal_001.tif # slide_id
βΒ Β βΒ Β βΒ Β Β Β βββ features.npy # features array (N_tiles_slide, 3+d)
...
βΒ Β Β Β βΒ Β βββ TCGA
βΒ Β Β Β βΒ Β β βββ TCGA_BRCA
βΒ Β Β Β βΒ Β β βΒ Β βββ TCGA-3C-AALI-01Z-00-DX1.F6E9A5DF-D8FB-45CF-B4BD-C6B76294C291.svs
βΒ Β βΒ Β βΒ Β βΒ Β Β Β βββ features.npy
...
βΒ Β Β Β βββ MoCoWideResNetCOAD # same structure applies for all extractors
βΒ Β Β Β βββ ResNet50
βΒ Β Β Β βββ iBOTViTBaseCOAD
βΒ Β Β Β βββ iBOTViTBasePANCAN
βΒ Β Β Β βββ iBOTViTLargeCOAD
βΒ Β Β Β βββ iBOTViTSmallCOAD
...
/!\ If you wish to extract features for Dino[ViT-S]BRCA, Dino[ViT-S]PanCancer, HIPT and CTransPath, those features should be placed here.
βΒ Β Β Β βββ DinoChenBRCA
βΒ Β Β Β βββ DinoChenPancancer
βΒ Β Β Β βββ HIPT
β βββ CTransPath
===============================================================================
βββ tiles_classification # tiles classification tasks
Β Β βββ features # features
Β Β βββ iBOTViTBasePANCAN # feature extractor
Β Β βΒ Β βββ CAMELYON17-WILDS_FULL # cohort
Β Β β βΒ Β βββ tile_features.npy # tiles features array (N_tiles_cohort, d)
Β Β β βΒ Β βββ tile_ids.npy # tiles ids array (N_tiles_cohort,)
Β Β βΒ Β βββ NCT-CRC_FULL
Β Β β Β Β βββ tile_features.npy
Β Β β Β Β βββ tile_ids.npy
Β Β Β Β βββ MoCoWideResNetCOAD
Β Β Β Β βββ ResNet50
Β Β Β Β βββ iBOTViTBaseCOAD
Β Β Β Β βββ iBOTViTBasePANCAN
Β Β Β Β βββ iBOTViTLargeCOAD
Β Β Β Β βββ iBOTViTSmallCOAD
You can find a detailed description of the repository in rl_benchmarks/README.md
file.
- Hardware
- Installation
- Feature extraction
- Slide-level downstream tasks
- Tile-level downstream tasks
- Notes
As a pre-requirement, we suggest to work on a machine with at least 8 CPUs, 32 Gb RAM and 1 GPU with at least 15Gb RAM. For instance, our experiments run on a Tesla T4 (15 Gb RAM), 32 Intel(R) Xeon(R) CPUs (@ 2.00GHz) and 64 Gb RAM.
rl_benchmarks
relies on the OpenSlide
library to read WSI. The python bindings are automatically installed with rl_benchmarks
library, but you will also need the C library:
- On Linux:
apt update && apt install openslide-tools
Pixman is a dependency of libopenslide
(the C library installed through apt
). Note that versions 0.3*
and 0.4*
gives different results one versus the other. Experiments were conducted with version 0.36.0
. You can change the system wide version of Pixman using apt
. The following command should show the version installed:
apt list --installed | grep pixman
If the returned version is not 0.36.0
you can try to install it with your package manager:
sudo apt update
sudo apt install libpixman-1-0
All the accessible versions are stored in this website.
You can eventually run apt-get check
to check for broken dependencies.
Create a dedicated conda environment (optional):
conda create -n rl_benchmarks python=3.8
conda activate rl_benchmarks
Install rl_benchmarks
package and its dependencies using the install.sh
file.
git clone https://github.com/owkin/HistoSSLscaling.git
cd ./HistoSSLscaling
# Install the RL_benchmarks repository (in editable mode) together with other requirements
python -m pip install -e . -r requirements.txt
Once the installation and data download steps are completed, you finally need to edit the conf.yaml
file so that to specify:
logs_save_dir
: directory for cross-validation experiments logsdata_dir
: root directory that contains the downloaded data. If you downloaded the data in/home/user/downloaded_data/
folder, then this should be thedata_dir
.
Once data has been downloaded and the previous installation steps done, you can run the full test suite to make sure features are loaded correctly. You first need to add specific requirements via:
python -m pip install -r requirements-tests.txt
Then, you can run the whole stack of tests by running the following command (within a tmux
session is strongly recommended):
bash dev_tools/run_tests.sh
You can also perform linting checks via:
bash dev_tools/linting.sh
[NOTE!] If you also wish to only test that your raw data (WSIs datasets and tiles datasets) follow the good structure, please run
pytest -v tests/ -m test_raw_data_loading
This repository enables you to extract and store the features associated with our iBOT[ViT-B]PanCancer
. Beforehand, you will need to download raw slides and strictly stick to the architecture described in the Data structure
section (raw
folder).
Note
If you are only interested in reproducing the results by running cross-validations, you can directly download and use coordinates and features (provided as numpy arra`ys) for all representation learning models and cohorts used in our publication.
To extract features for each slide of a slide-level dataset, use the following tool: ./tools/extract_features/extract_slide_features.py
.
python ./tools/extract_features/extract_slide_features.py \
feature_extractor=$feature_extractor \
slide_dataset=$dataset \
n_tiles=$n_tiles \
batch_size=$batch_size \
random_sampling=$random_sampling \
seed=$seed \
num_workers=$num_workers \
device=$device \
features_output_dir=$output_dir
Example:
python ./tools/extract_featuresextract_slide_features.py \
feature_extractor="iBOTViTBasePANCAN" \
slide_dataset="tcga_coad" \
n_tiles=1_000 \
batch_size=64 \
random_sampling=True \
seed=0 \
num_workers=8 \
device="[0,1]" \
features_output_dir=null
The following command extracts features from TCGA-COAD
cohort using our ViT-based iBOT model iBOT[ViT-B]PanCancer
. 1,000 slides per slide are extracted in a random order (with seed set to 0). Process uses 2 GPUs (id 0 and 1) and 8 workers. features_output_dir=null
will assign None
value to features_output_dir
. In that case, the path to the features output directory will automatically be picked up in conf.yaml
file.
Note
Slide features are saved as follows:
{features_path}/{feature_extractor}/{slide_dataset}/{slide_id}.{slide_format}/features.npy
For each slide, a (N_tiles, 3+d) numpy arrays is saved, with d
being the model's last layer. The 3 first columns rows correspond to (tile_level, x_coordinate, y_coordinate)
where tile_level
is computed with openslide.deepzoom.DeepZoomGenerator
(see "Data structure" section).
For example:
/workspace/data/preprocessed/slides_classification/features/ResNet50/TCGA/TCGA_COAD/TCGA-AA-3864-01Z-00-DX1.f6992bc7-ba05-4c30-9500-8f7b07b30f9a.svs/features.npy
To import them, you can use np.load
:
import numpy as np
features = np.load(βfeatures.npyβ)
assert features.shape == (n_tiles, feature_dim+3)
Warning
Once you have downloaded the data, tile levels and coordinates are automatically retrieved for each cohort and feature extractor. Our repository allows to generate the features used in our experiments. If you wish to change the tiles coordinates and level, you can create new coords.npy
files and change the path to coordinates folder in the constants.py
file.
To extract features for tile-level datasets (i.e. NCT-CRC and Camelyon17-WILDS), use the following tool: ./tools/extract_features/extract_tile_features.py
.
python ./tools/extract_features/extract_tile_features.py \
tile_dataset=$dataset \
feature_extractor=$feature_extractor \
batch_size=$batch_size \
seed=$seed \
num_workers=$num_workers \
device=$device \
output_dir=$output_dir
Example:
The following command extracts features from TCGA-COAD
cohort using our ViT-based iBOT model iBOT[ViT-B]PanCancer
. 1,000 slides per slide are extracted in a random order (with seed set to 0). Process uses 2 GPUs (id 0 and 1) and 8 workers. features_output_dir=null
will assign None
value to features_output_dir
. In that case, the path to the features output directory will automatically be picked up in conf.yaml
file.
python ./tools/extract_features/extract_tile_features.py \
tile_dataset="nct_crc" \
feature_extractor="iBOTViTBasePANCAN" \
batch_size=64 \
seed=0 \
num_workers=8 \
device="[0,1]" \
output_dir=null
Note
Tile features are saved as two numpy arrays, one containing the tile features (tile_features.npy
) and the other containing the corresponding tile ids (tile_ids.npy
) in {features_path}/{feature_extractor}/{tile_dataset}/
folder.
For example:
/workspace/data/preprocessed/tiles_classification/features/ResNet50/NCT-CRC_FULL/tile_features.npy
and /workspace/data/preprocessed/tiles_classification/features/ResNet50/NCT-CRC_FULL/tile_ids.npy
import numpy as np
features = np.load("tile_features.npy")
ids = np.load("tile_ids.npy")
assert features.shape == (n_samples, feature_dim)
assert ids.shape == (n_samples,)
Warning
If you wish to run all feature extractions sequentially, you can directly run
bash scripts/extract_slide_features.sh
bash scripts/extract_tile_features.sh
In those files, datasets and feature extractor are referenced as follows:
datasets="camelyon16_full tcga_coad tcga_kich tcga_kirc tcga_kirp tcga_luad tcga_lusc tcga_ov tcga_paad tcga_read tcga_stad "
feature_extractors="iBOTViTBasePANCAN"
If you wish to extract features (both at the slide and tile-level) for CTransPath
[1], HIPT
[2], DinoChenBRCA
[3] and DinoChenPancancer
[2], please directly use the corresponding repositories. Those models correspond to CTransPath
, HIPT
, Dino[ViT-S]BRCA
and HIPT[ViT_256]
:
CTransPath
([1], namedCTransPath
in our repository): extraction script. Wang, Xiyue, et al. "Transformer-based unsupervised contrastive learning for histopathological image classification." Medical image analysis 81 (2022): 102559.HIPT
([2], namedHIPT
in our repository): extraction script. Richard J. Chen, Chengkuan Chen, Yicong Li, Tiffany Y. Chen, Andrew D. Trister, Rahul G. Krishnan, and Faisal Mahmood. "Scaling vision transformers to gigapixel images via hierarchical self-supervised learning". In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16144β16155, June 2022.Dino[ViT-S]BRCA
([3], namedDinoChenBRCA
in our repository): extraction cript. Richard J Chen and Rahul G Krishnan. "Self-supervised vision transformers learn visual concepts in histopathology". Learning Meaningful Representations of Life, NeurIPS 2021, 2021.HIPT[ViT_256]
([23] Dino[ViT-S]PanCancer in our repository): weights.
After preprocessing of WSIs, we suggest using the above SSL models on tiles with same coordinates as provided in coords_256
(Dino[ViT-S]BRCA
, HIPT[ViT_256]
), coords_4096
(HIPT) and coords_224
(CTransPath). Generated features should follow, for each dataset, the same structure as described in the previous sections (1 features matrix for slides with (deepzoom_level, x, y) coordinates as first 3 columns, 1 features matrix for tiles-datasets).
This section describes how to run cross-validation experiments.
The scripts/slides_classification.sh
script allows you to run 5x5 nested cross-validations on slide classification tasks (TCGA cohorts and Camelyon16 dataset). No parameter tweaking should be performed. scripts/slides_classification.sh
iterates on:
- Slides classification tasks:
"
camelyon16_train_tumor_prediction
tcga_crc_msi_prediction
tcga_stad_msi_prediction
tcga_ov_hrd_prediction
tcga_brca_hrd_prediction
tcga_brca_histological_subtype_prediction
tcga_brca_molecular_subtype_prediction
tcga_nsclc_cancer_subtype_prediction
tcga_rcc_cancer_subtype_prediction
tcga_brca_os_prediction
tcga_coad_os_prediction
tcga_luad_os_prediction
tcga_lusc_os_prediction
tcga_paad_os_prediction
"
- Feature extractors (a thorough description is provided in our publication):
"
ResNet50
MoCoWideResNetCOAD
iBOTViTSmallCOAD
iBOTViTBaseCOAD
iBOTViTBasePANCAN
iBOTViTLargeCOAD
"
- Multiple Instance Learning models (description below):
"
mean_pool
chowder
abmil
dsmil
hipt_mil
trans_mil
"
HIPTMIL
is a lightweight transformer MIL aggregator used in Chen et al., page 6., to aggregate 4096x4096 features into a WSI-wise representation, further used for fine-tuning. The authors also call this MIL model 3-stages HIPT model with local and global pretraining. In this repo, we only consider frozen local and global transformers. Fine-tuning is only performed for the last transformer. See details here.DSMIL
is a MIL aggregator proposed in Li et al.. It takes tile-level features as input and produces a classification score based on a dual-stream mechanism (two branches). The first branch computes a "critical" instance (i.e. an important tile for the classification) and its score. The second branch computes a representation and an attention score for every tile (with respect to the critical instance), averages these representations to get a single slide-wise representation, and then computes a score. The final score is the average of the scores from the two branches.ABMIL
is a MIL aggregator proposed in Ilse et al.. It computes a representation and an attention score for every tile, and computes a slide-wise representation by averaging the representations with respect to the attention scores. This slide-wise representation is then passed to an MLP for the final task.Chowder
has been proposed in Courtiol et al.. It computes a score for each tile and selects only the top and bottomN
scores. These2N
scores are then passed to an MLP for the downstream task.MeanPool
computes the slide-average representation from all tiles and applies an MLP on top of it.TransMIL
implements the model proposed by Shao et al.. The TransMIL model is composed of the following steps: 1) sequence squaring, 2) Correlation modelling, 3) Position encoding (Pyramid Position Encoding Generator), 4) Deep Feature Aggregation and 5) Classification (see Figure 3 and Algorithm 2 in Shao et al.).
During nested-cross validations, gridsearching is performed on 2 hyperparameters: learning rate (
learning_rate_gs="[1.0e-3,1.0e-4]"
weight_decay_gs="[0.,1.0e-4]"
Also, stratification is performed at the patient level:
stratified=True
split_mode="patient_split"
The script tools/slide_level_tasks/get_results.py
allows you to retrieve slide-classification results from each experiments. Output results take the form of a pd.DataFrame
with all experiments' parameters and corresponding results. To get the results of nested cross-validations, simply do:
python tools/slide_level_tasks/get_results.py
Warning
hipt_mil
algorithm needs to set 1 slide's feature matrix per batch (batch size = 1). You can find the original implementation by HIPT authors here.
The scripts/tiles_classification.sh
script allows you to run cross-validations and test evaluation on tile classification tasks on NCT-CRC and Camelyon17-WILDS datasets using a standard SGD classifier on top of frozen features. This script makes use of tools/tile_level_tasks/linear_evaluation.py
which performs linear evaluation and stores metrics accordingly.
scripts/slides_classification.sh
iterates on:
- tiles classification tasks:
"camelyon17_wilds nct_crc"
- feature extractors:
"
ResNet50
MoCoWideResNetCOAD
iBOTViTSmallCOAD
iBOTViTBaseCOAD
iBOTViTBasePANCAN
iBOTViTLargeCOAD
"
Note
Once corresponding features have been extracted and stored appropriately according to our data structure (see first section), you can run the above experiments on CTransPath
, HIPT
, DinoChenBRCA
and DinoChenPancancer
by simply adding in the feature_extractors
parameter (bash scripts): CTransPath HIPT DinoChenBRCA DinoChenPancancer
. Note that should not use HIPT for tiles classification but rather the first ViT-S/256 extractors (which is denoted by DinoChenPancancer
).
When loading iBOTViTBasePANCAN
, you may encounter the following message:
Pretrained weights found at <your_data_dir>/weights/ibot_vit_base_pancan.pth and loaded with msg: _IncompatibleKeys(missing_keys=[], unexpected_keys=['head.mlp.0.weight', 'head.mlp.0.bias', 'head.mlp.2.weight', 'head.mlp.2.bias', 'head.mlp.4.weight', 'head.mlp.4.bias', 'head.last_layer.weight_g', 'head.last_layer.weight_v', 'head.last_layer2.weight_g', 'head.last_layer2.weight_v'])
If so, this message is normal as our weights also contain the final MLP head, which are is needed for features extraction.
- Add CI configuration in
.github/workflows/
. - Add Sphinx documentation.
Please open new issues directly on the repository, we'll do our best to address those quickly.
Vision Transformers architectures were derived from facebookresearch/dino (Apache License 2.0), huggingface/pytorch-image-models (Apache License 2.0) and lmlpen/Nystromformer (MIT License) repositories.
hipt_mil
multiple-instance learning algorithm was directly inspired from the HIPT repository (Apache License 2.0 with Commons Clause).
The following table describe the different libraries used in this work.
Name of the code library | Version | License | Licensor | Github repository |
---|---|---|---|---|
HIPT | - | Apache License 2.0 with Commons Clause | Mahmood Lab | https://github.com/mahmoodlab/HIPT/ |
dino | - | Apache License 2.0 | Not specified | https://github.com/facebookresearch/dino/ |
pytorch-image-models | 0.9.0 | Apache License 2.0 | Ross Wightman | https://github.com/huggingface/pytorch-image-models/ |
nystrom-attention | 0.0.11 | MIT License | Phil Wang | https://github.com/lucidrains/nystrom-attention/ |
einops | 0.6.1 | MIT License | Alex Rogozhnikov | https://github.com/arogozhnikov/einops/ |
hydra-core | 1.3.2 | MIT License | Facebook, Inc. and its affiliates | https://github.com/facebookresearch/hydra/ |
imageio | 2.31.1 | BSD-2 Clause | Imageio developers | https://github.com/imageio/imageio/ |
lifelines | 0.27.7 | MIT License | Cameron Davidson-Pilon | https://github.com/CamDavidsonPilon/lifelines/ |
loguru | 0.7.0 | MIT License | Not specified | https://github.com/Delgan/loguru/ |
openslide-python | 1.3.0 | GNU LGPL v2.1 | Free Software Foundation | https://github.com/openslide/openslide-python/ |
PyYAML | 6.0.1 | MIT License | Ingy dΓΆt Net and Kirill Simonov | https://github.com/yaml/pyyaml/ |
scikit-learn | 1.3.0 | BSD-3 Clause | Scikit-learn developers | https://github.com/scikit-learn/scikit-learn/ |
torch | 1.13.1 | Modified BSD Clause | See LICENSE | https://github.com/pytorch/pytorch/ |
torchvision | 0.14.1 | BSD-3 Clause | Soumith Chintala | https://github.com/pytorch/vision |
tqdm | 4.66.1 | MIT and Mozilla Public License | See LICENSE | https://github.com/tqdm/tqdm/ |
dill | 0.37.1 | BSD-3 Clause | The Uncertainty Quantification Foundation | https://github.com/uqfoundation/dill/ |
The following table describe the different datasets from which either features or labels were extracted.
Name of the dataset | License | Dataset home page |
---|---|---|
NCT-CRC-HE-100K | CC-BY 4.0 License | https://zenodo.org/record/1214456 |
Camelyon16 | CC0 1.0 License | https://camelyon17.grand-challenge.org/Data/ |
Camelyon17-WILDS | CC0 1.0 License | https://wilds.stanford.edu/datasets/#camelyon17 |
The results published here are also partly based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.
If you found our work useful in your research, please consider citing it at:
@article{Filiot2023ScalingSSLforHistoWithMIM,
author = {Alexandre Filiot and Ridouane Ghermi and Antoine Olivier and Paul Jacob and Lucas Fidon and Alice Mac Kain and Charlie Saillard and Jean-Baptiste Schiratti},
title = {Scaling Self-Supervised Learning for Histopathology with Masked Image Modeling},
elocation-id = {2023.07.21.23292757},
year = {2023},
doi = {10.1101/2023.07.21.23292757},
publisher = {Cold Spring Harbor Laboratory Press},
url = {https://www.medrxiv.org/content/early/2023/07/26/2023.07.21.23292757},
eprint = {https://www.medrxiv.org/content/early/2023/07/26/2023.07.21.23292757.full.pdf},
journal = {medRxiv}
}