medmnistc 0.1.0

This Python library aims to evaluate model robustness under corrupted test sets and to enhance domain generalization through domain-specific augmentations.

🏥 MedMNIST-C

We introduce MedMNIST-C [preprint], a benchmark dataset based on the MedMNIST+ collection covering 12 2D datasets and 9 imaging modalities. We simulate task and modality-specific image corruptions of varying severity to comprehensively evaluate the robustness of established algorithms against real-world artifacts and distribution shifts. We further show that our simple-to-use artificial corruptions allow for highly performant, lightweight data augmentation to enhance model robustness.

Installation and Requirements

pip install medmnistc

We do require Wand for image manipulation, a Python binding for ImageMagick. Thus, if you are using Ubuntu:

sudo apt-get install libmagickwand-dev

otherwise, please check the tutorial.

Main components

• medmnistc/corruptions/registry.py: List of all the corruptions and respective intensity hyperparameters.
• medmnistc/dataset_manager.py: Dataset class responsible for the creation of the corrupted datasets.
• medmnistc/visualizer.py: Class used to visualize and store the defined corruptions.
• medmnistc/augmentation.py: Augumentation class based on the defined corruptions.
• medmnistc/dataset.py: Dataset class used for the corrupted datasets.
• medmnistc/eval.py: PyTorch class used for model evaluation under corrupted datasets.
• medmnistc/assets/baseline/*: Normalization baselines used for model evaluation under corrupted datasets.

Basic usage

Create the corrupted datasets

from medmnistc.dataset_manager import DatasetManager

medmnist_path = ... # PATH TO THE CLEAN IMAGES
medmnistc_path = ... # PATH TO THE CORRUPTED IMAGES

ds_manager = DatasetManager(medmnist_path = medmnist_path, output_path=output_path)
ds_manager.create_dataset(dataset_name = "breastmnist") # create a single corrupted test set
ds_manager.create_dataset(dataset_name = "all") # create all

Augmentations

from medmnistc.augmentation import AugMedMNISTC
from medmnistc.corruptions.registry import CORRUPTIONS_DS
import torchvision.transforms as transforms

dataset = "breastmnist" # select dataset
train_corruptions = CORRUPTIONS_DS[dataset] # load the designed corruptions for this dataset
images = ... # load images

# Augment with AugMedMNISTC
augment = AugMedMNISTC(train_corruptions)
augmented_img = augment(images[0])

# Integrate into transforms.Compose
aug_compose = transforms.Compose([
    AugMedMNISTC(train_corruptions),
    transforms.ToTensor(),
    transforms.Normalize(mean=..., std=...)
])

augmented_img = aug_compose(images[0])

Notebooks

• Create the dataset
• Visualize the corruptions
• Evaluate the corruptions
• Use the designed augmentations

License

The code is under Apache-2.0 License.

The MedMNIST-C dataset is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), except DermaMNIST-C under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).

Citation

If you find this work useful, please consider citing us:

@article{,
   title={},
   author={},
   journal={},
   year={}
}

DISCLAIMER: This repository is inspired by MedMNIST APIs and the ImageNet-C repository. Thus, please also consider citing MedMNIST, the respective source datasets (described here) and ImageNet-C.

Release versions

• v0.1.0: MedMNIST-C beta release.