Dataset Viewer
Auto-converted to Parquet Duplicate
image_id
string
image
image
patient_id
string
pathology
string
birads
string
label
int32
bus_0001-l
1
malignant
4
1
bus_0001-r
1
malignant
4
1
bus_0002-l
2
benign
4
0
bus_0002-r
2
benign
4
0
bus_0003-l
3
malignant
4
1
bus_0003-r
3
malignant
4
1
bus_0004-l
4
benign
4
0
bus_0004-r
4
benign
4
0
bus_0005-l
5
malignant
4
1
bus_0005-r
5
malignant
4
1
bus_0006-s
6
malignant
4
1
bus_0007-l
7
benign
4
0
bus_0007-r
7
benign
4
0
bus_0008-l
8
benign
4
0
bus_0008-r
8
benign
4
0
bus_0009-l
9
malignant
4
1
bus_0009-r
9
malignant
4
1
bus_0010-l
10
malignant
4
1
bus_0010-r
10
malignant
4
1
bus_0011-l
11
benign
4
0
bus_0011-r
11
benign
4
0
bus_0012-l
12
malignant
4
1
bus_0012-r
12
malignant
4
1
bus_0013-l
13
benign
4
0
bus_0013-r
13
benign
4
0
bus_0014-s
14
malignant
4
1
bus_0015-s
15
malignant
4
1
bus_0016-l
16
malignant
4
1
bus_0016-r
16
malignant
4
1
bus_0017-s
17
malignant
4
1
bus_0018-s
18
malignant
4
1
bus_0019-l
19
benign
4
0
bus_0019-r
19
benign
4
0
bus_0020-l
20
benign
4
0
bus_0020-r
20
benign
4
0
bus_0021-l
21
malignant
4
1
bus_0021-r
21
malignant
4
1
bus_0022-l
22
benign
4
0
bus_0022-r
22
benign
4
0
bus_0023-l
23
benign
4
0
bus_0023-r
23
benign
4
0
bus_0024-l
24
benign
4
0
bus_0024-r
24
benign
4
0
bus_0025-l
25
benign
4
0
bus_0025-r
25
benign
4
0
bus_0026-s
26
benign
4
0
bus_0027-l
27
malignant
4
1
bus_0027-r
27
malignant
4
1
bus_0028-l
28
benign
4
0
bus_0028-r
28
benign
4
0
bus_0029-s
29
malignant
4
1
bus_0030-s
30
malignant
4
1
bus_0031-s
31
malignant
4
1
bus_0032-l
32
malignant
4
1
bus_0032-r
32
malignant
4
1
bus_0033-l
33
benign
4
0
bus_0033-r
33
benign
4
0
bus_0034-l
34
malignant
4
1
bus_0034-r
34
malignant
4
1
bus_0035-l
35
benign
4
0
bus_0035-r
35
benign
4
0
bus_0036-l
36
benign
4
0
bus_0036-r
36
benign
4
0
bus_0037-s
37
malignant
4
1
bus_0038-s
38
benign
4
0
bus_0039-l
39
benign
4
0
bus_0039-r
39
benign
4
0
bus_0040-s
40
benign
4
0
bus_0041-l
41
malignant
4
1
bus_0041-r
41
malignant
4
1
bus_0042-l
42
benign
4
0
bus_0042-r
42
benign
4
0
bus_0043-s
43
malignant
4
1
bus_0044-l
44
benign
4
0
bus_0044-r
44
benign
4
0
bus_0045-l
45
malignant
4
1
bus_0045-r
45
malignant
4
1
bus_0046-l
46
malignant
4
1
bus_0046-r
46
malignant
4
1
bus_0047-l
47
benign
2
0
bus_0047-r
47
benign
2
0
bus_0048-l
48
malignant
4
1
bus_0048-r
48
malignant
4
1
bus_0049-l
49
malignant
4
1
bus_0049-r
49
malignant
4
1
bus_0050-l
50
malignant
4
1
bus_0050-r
50
malignant
4
1
bus_0051-l
51
benign
4
0
bus_0051-r
51
benign
4
0
bus_0052-l
52
malignant
4
1
bus_0052-r
52
malignant
4
1
bus_0053-l
53
benign
4
0
bus_0053-r
53
benign
4
0
bus_0054-s
54
benign
4
0
bus_0055-l
55
benign
4
0
bus_0055-r
55
benign
4
0
bus_0056-l
56
benign
4
0
bus_0056-r
56
benign
4
0
bus_0057-l
57
malignant
4
1
bus_0057-r
57
malignant
4
1
End of preview. Expand in Data Studio

BUS-BRA โ€” Breast Ultrasound Images for Computer-Aided Diagnosis

This dataset card was generated with the help of generative AI and checked against the original sources. If you find an error or anything unclear, please open a discussion in the Community tab of this repository.

This dataset is derived from BUS-BRA (Gรณmez-Flores et al., 2024), a publicly available dataset of anonymized breast ultrasound (BUS) images of 1,064 patients who underwent routine breast studies, originally hosted on Zenodo. The dataset contains biopsy-proven tumor cases with BI-RADS annotations in categories 2, 3, 4 and 5. We kept the original PNG images unchanged and added the patient ID, the pathology, the BI-RADS category and an integer label from the original metadata file. The dataset is designed for benign vs. malignant breast tumor classification.

๐Ÿ“ฆ Dataset Structure

The dataset has a single train split, with each row corresponding to a single ultrasound image:

Field Description
image_id Original image ID (e.g., bus_0001-l)
image Breast ultrasound image (variable size, PNG)
patient_id Patient (case) number from the original metadata
pathology benign or malignant
birads BI-RADS category: 2, 3, 4 or 5
label Integer pathology: 0 = benign, 1 = malignant

In total, the dataset contains 1,875 images from 1,064 patients.

Class Distribution:

Pathology BI-RADS 2 BI-RADS 3 BI-RADS 4 BI-RADS 5 Total
Benign 552 432 280 4 1,268
Malignant 10 31 413 153 607
Total 562 463 693 157 1,875

โš™๏ธ Preprocessing

  • BUSBRA.zip was downloaded from Zenodo (record 8231412, version 1.0)
  • Images were kept as the original PNG files (no re-encoding, resizing or cropping)
  • Each image was matched to its row in the original metadata file by its ID
  • patient_id, pathology and birads were taken from the Case, Pathology and BIRADS columns
  • label is 1 for malignant and 0 for benign
  • The tumor delineation masks of the original release were not included
  • All images listed in the metadata file were preserved

๐Ÿš€ Usage

from datasets import load_dataset
import matplotlib.pyplot as plt
from sklearn.model_selection import GroupShuffleSplit

ds = load_dataset("chehablab/BUS-BRA", split="train")

# One benign and one malignant example
labels = ds["label"]
fig, axes = plt.subplots(1, 2, figsize=(8, 4))
for ax, label in zip(axes, [0, 1]):
    sample = ds[labels.index(label)]
    ax.imshow(sample["image"], cmap="gray")
    ax.set_title(f"{sample['pathology']} | BI-RADS {sample['birads']}")
    ax.axis("off")
plt.tight_layout()
plt.show()

# No official split: split by patient so no patient appears in both sets
gss = GroupShuffleSplit(n_splits=1, test_size=0.2, random_state=0)
train_idx, test_idx = next(gss.split(range(len(ds)), groups=ds["patient_id"]))
train, test = ds.select(train_idx), ds.select(test_idx)

๐Ÿท๏ธ Feature Descriptions

Pathology

  • Benign (label=0): 1,268 images
  • Malignant (label=1): 607 images

BI-RADS

The Breast Imaging Reporting and Data System (BI-RADS) category assigned to each image:

  • BI-RADS 2 โ€” Benign finding
  • BI-RADS 3 โ€” Probably benign
  • BI-RADS 4 โ€” Suspicious for malignancy
  • BI-RADS 5 โ€” Highly suggestive of malignancy

Image Specifications

  • Modality: 2D breast ultrasound (B-mode)
  • Image size: Variable, PNG
  • Content: One breast lesion per image

Acquisition Protocol

  • Source: Routine breast studies of 1,064 patients
  • Copyright and distribution: Program of Biomedical Engineering of the Federal University of Rio de Janeiro (PEB/COPPE-UFRJ), Brazil, with the Centro de Investigaciรณn y de Estudios Avanzados (Cinvestav), Mexico

๐Ÿงช Use Cases

  • Benign vs. malignant breast tumor classification
  • BI-RADS category prediction
  • Studying agreement between BI-RADS assessment and pathology
  • Representation learning and feature extraction for breast ultrasound

โš ๏ธ Important Notes

  • No official split: The dataset has a single train split. Make your own split by patient_id (see Usage), since most patients have more than one image.
  • Class imbalance: Benign images (1,268) are about twice as frequent as malignant images (607). Use weighted sampling or class balancing as needed.
  • BI-RADS vs. pathology: BI-RADS is the radiologist's assessment and does not always match the pathology (e.g., 10 malignant images are BI-RADS 2). Use pathology as the ground truth for benign vs. malignant.
  • Masks not included: The original release also provides tumor delineations; see the original Zenodo record for segmentation tasks.

๐Ÿ“š Citation

This dataset is derived from data originally published on Zenodo under a CC BY 4.0 license. The original authors require any research using the dataset to cite their paper. If you use this dataset, please acknowledge Chehab Lab and cite the original BUS-BRA dataset:

@misc{GomezFlores2023BUSBRAData,
  author    = {G{\'o}mez-Flores, Wilfrido and Gregorio-Calas, Maria Julia and Pereira, Wagner Coelho de Albuquerque},
  title     = {BUS-BRA: A Breast Ultrasound Dataset for Assessing Computer-aided Diagnosis Systems [dataset]},
  year      = {2023},
  publisher = {Zenodo},
  version   = {1.0},
  doi       = {10.5281/zenodo.8231412}
}

Additionally, cite the research paper that introduced this dataset:

@article{GomezFlores2024BUSBRA,
  author  = {G{\'o}mez-Flores, Wilfrido and Gregorio-Calas, Maria Julia and Pereira, Wagner Coelho de Albuquerque},
  title   = {BUS-BRA: A Breast Ultrasound Dataset for Assessing Computer-aided Diagnosis Systems},
  journal = {Medical Physics},
  volume  = {51},
  pages   = {3110--3123},
  year    = {2024},
  doi     = {10.1002/mp.16812}
}

โœ‰๏ธ Original Authors

Wilfrido Gรณmez-Flores, Maria Julia Gregorio-Calas, Wagner Coelho de Albuquerque Pereira
Program of Biomedical Engineering (PEB/COPPE), Federal University of Rio de Janeiro, Brazil, and Centro de Investigaciรณn y de Estudios Avanzados (Cinvestav), Mexico

Zenodo: https://zenodo.org/records/8231412 (original files)

๐Ÿ“œ License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, consistent with the original BUS-BRA dataset.

You may copy, modify, distribute, and use the data for any purpose (commercial or non-commercial), provided that appropriate credit is given to Gรณmez-Flores et al., Zenodo, and the original paper.

CC BY 4.0

Chehab Lab @ 2026

Downloads last month
-