BUS-UCLM — Breast Ultrasound Lesion Segmentation Dataset

Re-hosted mirror of the BUS-UCLM dataset (Vallez et al., Scientific Data 2025), collected at Hospital General de Ciudad Real, Spain on a Siemens ACUSON S2000 with the 18L6 HD probe. Originally distributed via Mendeley Data (DOI: 10.17632/7fvgj4jsp7.3).

This mirror is intended for use with EasyMedSeg and provides:

• A single canonical source (the original Mendeley S3 cache returns 403 to direct fetches; we mirror the canonical images/ + masks/ + INFO.csv).
• Parquet schema with one row per image and a single binary mask.
• The original mask format is RGB-color-coded by class (green for benign, red for malignant, all-zero for normal). This mirror collapses each mask to a binary L-mode image (any non-zero pixel → foreground). The class label is preserved in the class_label column.

Composition

Class	Images
benign	174
malignant	90
normal	419
Total	683

683 images from 38 patients (38 distinct patient prefixes in the filenames). Standard image dimensions ~856×606 px (varies by case; full per-image resolution is preserved in the resolution column).

Splits

BUS-UCLM has no official train/val/test split. This release ships a single train split. The authors' helper repo suggests a 33/5 patient partition with five test prefixes (COPE, ANFO, ELCO, CRCI, FLKA); reproduce it downstream with a patient_id-based filter if needed.

Schema

Column	Type	Description
image	Image	Source PNG (RGB)
mask	Image	Binary mask (L mode, 0/255) — collapsed from RGB-coded source
image_id	string	e.g., "ALWI_000"
patient_id	string	First underscore token (e.g., "ALWI")
class_label	string	"benign" / "malignant" / "normal"
resolution	string	Native resolution string from INFO.csv (e.g., "856x606")
has_doppler	bool	Whether the image is a Doppler frame
has_marks	bool	Whether the source image carries on-screen annotations (calipers, arrows)
has_combined	bool	Whether the source image is a "combined" (montage) acquisition

License

CC BY 4.0, inherited from the upstream Mendeley release.

Citation

@article{vallez2025bus,
  title   = {BUS-UCLM: Breast ultrasound lesion segmentation dataset},
  author  = {Vallez, Noelia and Bueno, Gloria and Deniz, Oscar and Rienda, Miguel Angel and Pastor, Carlos},
  journal = {Scientific Data},
  volume  = {12},
  number  = {1},
  pages   = {242},
  year    = {2025},
  doi     = {10.1038/s41597-025-04562-3}
}