BreakHis

Overview

The Breast Cancer Histopathological Image Classification (BreakHis) dataset contains 7,909 microscopic images of breast tumour tissue, collected from hematoxylin and eosin stained biopsy slides of 82 patients at the P&D Laboratory, Brazil. Each slide was imaged at four magnifications (40X, 100X, 200X and 400X). Each image is an RGB PNG of 700x460 pixels (74 images are 700x456), labelled with one of eight tumour types, four benign and four malignant.

• License: CC BY 4.0 (stated on the homepage); cite the original paper when using the data

• Folds: the authors’ five train/test folds used in the paper, downloaded from the homepage. Each fold keeps all images of a slide (one patient ID and tumour type) in the same split. Select a fold with config_name="fold1" (the default) to "fold5":

Fold	Train images	Test images
fold1	5,005	2,904
fold2	5,506	2,403
fold3	5,332	2,577
fold4	5,211	2,698
fold5	4,826	3,083

Data Structure

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

Key	Type	Description
image	PIL.Image.Image	RGB histopathology image, 700x460 (74 images are 700x456)
label	int	Tumour type, 0-7 (see Classes); label < 4 is benign
malignant	int	0 (benign) or 1 (malignant), for the binary task
magnification	int	0 (40X), 1 (100X), 2 (200X) or 3 (400X)
patient_id	str	Patient ID from the file name, e.g. "14-22549AB"

Classes

Label	Tumour type	Benign/malignant	Train / test images (fold1)
0	adenosis	benign	253 / 191
1	fibroadenoma	benign	683 / 331
2	phyllodes_tumor	benign	218 / 235
3	tubular_adenoma	benign	318 / 251
4	ductal_carcinoma	malignant	2,313 / 1,138
5	lobular_carcinoma	malignant	302 / 324
6	mucinous_carcinoma	malignant	547 / 245
7	papillary_carcinoma	malignant	371 / 189

Usage Example

from stable_datasets.images import BreakHis

# Downloads the 4.3 GB archive on first use; config_name defaults to "fold1"
ds = BreakHis(split="train")
test_fold3 = BreakHis(config_name="fold3", split="test")
sample = ds[0]

print(sample.keys())          # dict_keys(["image", "label", "malignant", "magnification", "patient_id"])
print(sample["image"].size)   # (700, 460), or (700, 456) for 74 images
print(sample["label"])        # tumour type, 0-7

# Binary benign/malignant task at one magnification
ds_200x = ds.filter(lambda ex: ex["magnification"] == 2)
binary_label = ds_200x[0]["malignant"]

# Load as PyTorch tensors
ds_torch = ds.with_format("torch")
print(ds_torch[0]["image"].shape)  # 3 x 460 x 700

Citation

@article{spanhol2016dataset,
    title={A Dataset for Breast Cancer Histopathological Image Classification},
    author={Spanhol, Fabio A and Oliveira, Luiz S and Petitjean, Caroline and Heutte, Laurent},
    journal={IEEE Transactions on Biomedical Engineering},
    volume={63},
    number={7},
    pages={1455--1462},
    year={2016},
    publisher={IEEE}
}