Datasets:

Beothuk
/

dermamnist-federated

license: cc-by-nc-4.0
task_categories:
- image-classification
tags:
- medmnist
- medical-imaging
- skin-cancer
- federated-learning
- healthcare
size_categories:
- 10K<n<100K
---

# DermaMNIST Dataset for Federated Learning

## Dataset Description

**DermaMNIST** is a multi-class dataset of dermatoscopic images for skin lesion classification. It is part of the [MedMNIST v2](https://medmnist.com/) collection, a set of standardized biomedical image datasets for machine learning. This version of the dataset has been prepared for easy use in federated learning simulations, particularly with the [Flower](https://flower.dev/) framework.

The dataset is based on the [HAM10000](https://doi.org/10.1038/sdata.2018.161) dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. The images are 28x28 pixels and resized from their original 600x450 resolution.

### Classes
The dataset contains 7 classes of skin lesions:
- **0**: `actinic keratoses and intraepithelial carcinoma`

1: basal cell carcinoma
2: benign keratosis-like lesions
3: dermatofibroma
4: melanoma
5: melanocytic nevi

6: vascular lesions

Dataset Statistics

Number of classes: 7
Training samples: 7,007
Test samples: 2,005
Total samples: 9,012

Dataset Structure

Data Fields

image: A PIL Image object of size 28x28.
label: The integer class ID.
label_name: The string name of the class label.

Data Splits

Train: 7,007 samples
Test: 2,005 samples

Usage with Flower Datasets 🌸

This dataset is ideal for federated learning simulations where data is distributed across multiple clients (e.g., hospitals). Here's how to use it with Flower Datasets:

from flwr_datasets import FederatedDataset

# Define the number of clients (e.g., simulating 10 hospitals)
NUM_CLIENTS = 10

# Load the dataset and partition it into 10 clients
fds = FederatedDataset(dataset="your-username/dermamnist-fl", partitioners={"train": NUM_CLIENTS})

# Get the data for a specific client
client_data = fds.load_partition(0, "train")

# You can now use this client_data with your standard PyTorch or TensorFlow data loaders.
# For example, to set the format for PyTorch:
# client_data.set_format(type="torch", columns=["image", "label"])

Citation

If you use this dataset in your research, please cite the following paper:

@article{medmnistv2,
    title={MedMNIST v2: A Large-Scale Lightweight Benchmark for 2D and 3D Biomedical Image Classification},
    author={Jiancheng Yang and Rui Shi and Donglai Wei and Zequan Liu and Lin Zhao and Bilian Ke and Hanspeter Pfister and Bingbing Ni},
    journal={Scientific Data},
    year={2023},
    volume={10},
    number={1},
    pages={41},
    doi={10.1038/s41597-022-01721-8}
}

Citation

    @article{medmnistv2,
        title={MedMNIST v2: A Large-Scale Lightweight Benchmark for 2D and 3D Biomedical Image Classification},
        author={Jiancheng Yang and Rui Shi and Donglai Wei and Zequan Liu and Lin Zhao and Bilian Ke and Hanspeter Pfister and Bingbing Ni},
        journal={Scientific Data},
        year={2023},
        volume={10},
        number={1},
        pages={41},
        doi={10.1038/s41597-022-01721-8}
    }

Example Usage with datasets

from datasets import load_dataset
from PIL import Image

# Load the dataset
dataset = load_dataset("your-username/dermamnist-fl")

# Access the training split
train_data = dataset["train"]
print(f"Number of training samples: {len(train_data)}")

# Access a sample
sample = train_data[100]
image: Image = sample['image']
print(f"Image size: {image.size}")
print(f"Label: {sample['label']}")
print(f"Label name: {sample['label_name']}")

# To display the image
# image.show()