jusiro / FLAIR

[MedIA'24] FLAIR: A Foundation LAnguage-Image model of the Retina for fundus image understanding.

FLAIR: A Foundation LAnguage Image model of the Retina

A Foundation LAnguage Image model of the Retina (FLAIR):
Encoding expert knowledge in text supervision
Julio Silva-Rodriguez¹, Hadi Chakor², Riadh Kobbi², Jose Dolz¹, Ismail Ben Ayed^1
1LIVIA - ETS Montreal, ²DIAGNOS Inc.
| Project | Paper | Code | Tutorials |

Install FLAIR

• Install in your enviroment a compatible torch version with your GPU. For example:

conda create -n flair_env python=3.8 -y
conda activate flair_env
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge

• Install FLAIR library.

pip install git+https://github.com/jusiro/FLAIR.git

Usage

from PIL import Image
import numpy as np

# Import FLAIR
from flair import FLAIRModel

# Set model
model = FLAIRModel(from_checkpoint=True)

# Load image and set target categories 
# (if the repo is not cloned, download the image and change the path!)

image = np.array(Image.open("./documents/sample_macular_hole.png"))
text = ["normal", "healthy", "macular edema", "diabetic retinopathy", "glaucoma", "macular hole",
        "lesion", "lesion in the macula"]

# Forward FLAIR model to compute similarities
probs, logits = model(image, text)

print("Image-Text similarities:")
print(logits.round(3)) # [[-0.32  -2.782  3.164  4.388  5.919  6.639  6.579 10.478]]
print("Probabilities:")
print(probs.round(3))  # [[0.      0.     0.001  0.002  0.01   0.02   0.019  0.948]]

Note: problems during automatic pre-trained weights download

If you encounter any issue while downloading the pre-trained weights (i.e. from_checkpoint=True), you can manually download the weights from the following links (see Table), unzip the file, and store them at: ./flair/modeling/flair_pretrained_weights/[ID].pth.

Backbone	ID
ResNet-50	flair_resnet	LINK

Pre-training and transferability

In the following, we present the scripts for model pre-training and transferability. To use them, we recommend cloning the whole repository.

git clone https://github.com/jusiro/FLAIR.git
cd FLAIR
pip install -r requirements.txt

📦 Foundation model pre-training

• Define the relative paths for datasets and dataframes in ./local_data/constants.py.

• Prepare the FUNDUS assembly dataset - check ./local_data/prepare_partitions.py to prepare the dataframes.

01_EYEPACS	08_ODIR-5K	15_APTOS	22_HEI-MED	29_AIROGS	36_ACRIMA
02_MESIDOR	09_PAPILA	16_FUND-OCT	23_HRF	30_SUSTech-SYSU	37_DeepDRiD
03_IDRID	10_PARAGUAY	17_DiaRetDB1	24_ORIGA	31_JICHI
04_RFMid	11_STARE	18_DRIONS-DB	25_REFUGE	32_CHAKSU
05_1000x39	12_ARIA	19_Drishti-GS1	26_ROC	33_DR1-2
06_DEN	13_FIVES	20_E-ophta	27_BRSET	34_Cataract
07_LAG	14_AGAR300	21_G1020	28_OIA-DDR	35_ScarDat

• Vision-Language Pre-training.

python main_pretrain.py --augment_description True --balance True --epochs 15 --batch_size 128 --num_workers 6

📦 Transferability to downstream tasks/domains

• Define the relative paths for datasets and dataframes in ./local_data/constants.py.
• Prepare the experiment setting for the target dataset - we used ./local_data/experiments.py to store them.

if experiment == "02_MESSIDOR":
    setting = {"dataframe": PATH_DATAFRAME_TRANSFERABILITY_CLASSIFICATION + "02_MESSIDOR.csv",
               "task": "classification",
               "targets": {"no diabetic retinopathy": 0,
                           "mild diabetic retinopathy": 1,
                           "moderate diabetic retinopathy": 2,
                           "severe diabetic retinopathy": 3,
                           "proliferative diabetic retinopathy": 4}}

• Zero-shot (no adaptation).

python main_transferability.py --experiment 02_MESSIDOR --method zero_shot --load_weights True --domain_knowledge True  --shots_train 0% --shots_test 100% --project_features True --norm_features True --folds 1 

• Linear Probing.

python main_transferability.py --experiment 02_MESSIDOR --method lp --load_weights True --shots_train 80% --shots_test 20% --project_features False --norm_features False --folds 5 

Citation

If you find this repository useful, please consider citing this paper:

@article{FLAIR2023,
  title={A Foundation LAnguage-Image model of the Retina (FLAIR): Encoding expert knowledge in text supervision},
  author={Julio Silva-Rodriguez and Hadi Chakor and Riadh Kobbi and Jose Dolz and Ismail Ben Ayed},
  journal={ArXiv Preprint},
  year={2023}
}

License

• Code and Model Weights are released under Apache 2.0 license