adrianrm/breastmnist

---
license: cc-by-4.0
task_categories:
- image-classification
tags:
- medical
- medmnist
- breastmnist
configs:
- config_name: train-all-res224
data_files:
- split: train
path: train-all-res224/*.parquet
- config_name: train-malignant-res224
data_files:
- split: train
path: train-malignant-res224/*.parquet
- config_name: train-normal_benign-res224
data_files:
- split: train
path: train-normal_benign-res224/*.parquet
- config_name: val-all-res224
data_files:
- split: val
path: val-all-res224/*.parquet
- config_name: val-malignant-res224
data_files:
- split: val
path: val-malignant-res224/*.parquet
- config_name: val-normal_benign-res224
data_files:
- split: val
path: val-normal_benign-res224/*.parquet
- config_name: test-all-res224
data_files:
- split: test
path: test-all-res224/*.parquet
- config_name: test-malignant-res224
data_files:
- split: test
path: test-malignant-res224/*.parquet
- config_name: test-normal_benign-res224
data_files:
- split: test
path: test-normal_benign-res224/*.parquet
---

# breastmnist (MedMNIST)

**Source:** [breastmnist](https://medmnist.com/)
**Task:** binary-class
**Resolutions:** 224x224
**License:** CC BY 4.0

## Description

The BreastMNIST is based on a dataset of 780 breast ultrasound images. It is categorized into 3
classes: normal, benign, and malignant. As we use low-resolution images, we simplify the task into
binary classification by combining normal and benign as positive and classifying them against
malignant as negative. We split the source dataset with a ratio of 7:1:2 into training, validation
and test set. The source images of 1×500×500 are resized into 1×28×28.

## Config naming convention

```
{split}-{class}-{res}

split : train | val | test
class : all | <sanitized class name>
res : res28 | res64 | res128 | res224
```

## Loading examples

```python
from datasets import load_dataset

# All training images at 224px
ds = load_dataset('.../breastmnist', 'train-all-res224', split='train')

# Only 'malignant' class, training split
ds = load_dataset('.../breastmnist', 'train-malignant-res224', split='train')
```

## Class labels

- `0` — malignant (config key: `malignant`)
- `1` — normal, benign (config key: `normal_benign`)

## Class distribution

### 224x224

**train** (N=546, IR=2.71x)

| Class | Config key | Count | Share |
|-------|-----------|------:|------:|
| malignant | `malignant` | 147 | 26.9% |
| normal, benign | `normal_benign` | 399 | 73.1% |

**val** (N=78, IR=2.71x)

| Class | Config key | Count | Share |
|-------|-----------|------:|------:|
| malignant | `malignant` | 21 | 26.9% |
| normal, benign | `normal_benign` | 57 | 73.1% |

**test** (N=156, IR=2.71x)

| Class | Config key | Count | Share |
|-------|-----------|------:|------:|
| malignant | `malignant` | 42 | 26.9% |
| normal, benign | `normal_benign` | 114 | 73.1% |

## Citation

```bibtex
@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},
year={2023}
}
```