MedMNIST

Overview

MedMNIST is a large-scale MNIST-like collection of standardized biomedical images. In stable-datasets, MedMNIST is exposed via the MedMNIST class, but the actual dataset depends on the selected variant (passed as config_name, e.g. dermamnist, pathmnist). Each variant provides train/validation/test splits.

All 2D variants are pre-processed to 28×28 images and all 3D variants are pre-processed to 28×28×28 volumes, with corresponding labels.

Variants (2D)

2D variants use 28×28 images.

Variant	Classes	Train	Validation	Test	Notes
pathmnist	9	89,996	10,004	7,180
chestmnist	14	78,468	11,219	22,433	multi-label
dermamnist	7	7,007	1,003	2,005
octmnist	4	97,477	10,832	1,000
pneumoniamnist	2	4,708	524	624
retinamnist	5	1,080	120	400	ordinal regression
breastmnist	2	546	78	156
bloodmnist	8	11,959	1,712	3,421
tissuemnist	8	165,466	23,640	47,280
organamnist	11	34,561	6,491	17,778
organcmnist	11	12,975	2,392	8,216
organsmnist	11	13,932	2,452	8,827

Variants (3D)

3D variants use 28×28×28 volumes.

Variant	Classes	Train	Validation	Test
organmnist3d	11	971	161	610
nodulemnist3d	2	1,158	165	310
adrenalmnist3d	2	1,188	98	298
fracturemnist3d	3	1,027	103	240
vesselmnist3d	2	1,335	191	382
synapsemnist3d	2	1,230	177	352

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys.

2D variants

Key	Type	Description
image	PIL.Image.Image	28×28 image
label	int / list[int]	Class label (range depends on the selected variant; chestmnist uses a multi-label vector)

3D variants

Key	Type	Description
image	list	28×28×28 volume
label	int	Class label (range depends on the selected variant)

Usage Example

Basic Usage (2D variant)

from stable_datasets.images.med_mnist import MedMNIST

# Pick a 2D variant via config_name
variant = "dermamnist"

ds_train = MedMNIST(split="train", config_name=variant)
ds_val = MedMNIST(split="validation", config_name=variant)
ds_test = MedMNIST(split="test", config_name=variant)

sample = ds_train[0]
print(sample.keys())  # {"image", "label"}

image = sample["image"]  # PIL.Image.Image
label = sample["label"]  # int

# Optional: make it PyTorch-friendly
ds_train_torch = ds_train.with_format("torch")

Basic Usage (2D multi-label variant: chestmnist)

from stable_datasets.images.med_mnist import MedMNIST

variant = "chestmnist"
ds_train = MedMNIST(split="train", config_name=variant)

sample = ds_train[0]
image = sample["image"]  # PIL.Image.Image
label = sample["label"]  # multi-label vector (length 14)

# Example: indices of positive labels
positives = [i for i, v in enumerate(label) if int(v) == 1]
print("positive label indices:", positives)

Basic Usage (3D variant)

from stable_datasets.images.med_mnist import MedMNIST

variant = "organmnist3d"

ds_train = MedMNIST(split="train", config_name=variant)
sample = ds_train[0]

image = sample["image"]  # nested list, shape (28, 28, 28)
label = sample["label"]  # int

References

• Official website: https://medmnist.com/

• License: CC BY 4.0

Citation

@article{medmnistv2,
  title={MedMNIST v2-A large-scale lightweight benchmark for 2D and 3D biomedical image classification},
  author={Yang, Jiancheng and Shi, Rui and Wei, Donglai and Liu, Zequan and Zhao, Lin and Ke, Bilian and Pfister, Hanspeter and Ni, Bingbing},
  journal={Scientific Data},
  volume={10},
  number={1},
  pages={41},
  year={2023},
  publisher={Nature Publishing Group UK London}
}