BRACS

BRACS provides a large annotated Hematoxylin and Eosin (H&E)-stained whole-slide image (WSI) dataset with Regions Of Interest (ROIs) for training and evaluating deep learning models for breast lesion subtyping.

Key Features:

• Dataset composition: 547 Whole-Slide Images (WSIs) and 4,539 Regions Of Interest (ROIs) of H&E-stained breast tissue.
• Annotation process: Annotations produced through a consensus review involving three board-certified pathologists.
• Lesion categorization: Labels include three primary types—benign, malignant, and atypical—with further subdivision into seven distinct subtypes.
• Inclusion of atypical lesions: Atypical lesions are explicitly included to represent lesion heterogeneity and challenging diagnostic categories.
• Annotation levels: Provided annotations at both WSI and ROI levels for multi-scale analysis.
• Intended analytical use: Curated to support training, development, and evaluation of deep learning algorithms for automated pathology and breast cancer subtyping.

Scientific Applications:

• Deep learning training and evaluation: Use for supervised training and benchmarking of convolutional neural networks and other machine learning models on WSI- and ROI-level classification tasks.
• Subtype characterization: Development and validation of algorithms to distinguish benign, malignant, and atypical lesions and to resolve seven specific subtypes.
• Heterogeneity analysis: Study of intra- and inter-lesion heterogeneity, including challenging atypical cases, to improve diagnostic models.

Methodology:

Annotations were created by consensus of three board-certified pathologists; the resource consists of H&E-stained WSIs and annotated ROIs provided for training and evaluating deep learning models.