BC-MultiSet: A Multi-Task Dataset for Nucleus Segmentation, Classification, and Receptor Status Prediction in Invasive Breast Carcinoma

BC-MultiSet is a multi-task histopathology dataset for invasive breast carcinoma, comprising 5,586 H&E-stained image patches (256 × 256 px, 40× magnification) derived from whole-slide images of 50 patients. Each patch includes per-nucleus polygon annotations for three cell categories: Tumor, Fibroblast, and Inflammatory (lymphocytes, plasma cells, macrophages, mast cells, neutrophils, eosinophils) independently labeled by two board-certified pathologists and reconciled by consensus, totaling 109,180 annotated nuclei. Every patient is further characterized by clinical and molecular data, including ER, PR, and HER2 status (raw and binarized), molecular subtype (Luminal vs Non-Luminal), lymph node involvement, Ki67 proliferation index, tumor-infiltrating lymphocyte (TILs) density, tumor grade, and age. Data are organized into four cross-validation folds, each with images, nucleus annotation files (.npy), and a types.csv metadata file, accompanied by precomputed class weights and a patient-level clinical spreadsheet. BC-MultiSet supports the development of multi-task models linking nucleus-level morphology to patient-level molecular and receptor status prediction in breast cancer. Dataset Structure BC-MultiSet-50 ├── fold1 │ ├── images/ 1,500 patches, 256 × 256 px (.png) │ ├── labels/ 1,500 nucleus annotation files (.npy) – Tumor, Inflammatory, Fibroblast │ └── types.csv ├── fold2 │ ├── images/ 1,386 patches, 256 × 256 px (.png) │ ├── labels/ 1,386 nucleus annotation files (.npy) – Tumor, Inflammatory, Fibroblast │ └── types.csv ├── fold3 │ ├── images/ 1,554 patches, 256 × 256 px (.png) │ ├── labels/ 1,554 nucleus annotation files (.npy) – Tumor, Inflammatory, Fibroblast │ └── types.csv ├── fold4 │ ├── images/ 1,151 patches, 256 × 256 px (.png) │ ├── labels/ 1,151 nucleus annotation files (.npy) – Tumor, Inflammatory, Fibroblast │ └── types.csv ├── class_weights.txt └── BC-MultiSet.xlsx Each labels/*.npy file stores per-nucleus polygon annotations for the corresponding image patch, with class membership restricted to Tumor, Inflammatory, or Fibroblast. The types.csv file in each fold maps patches to their respective tissue/type metadata, class_weights.txt provides precomputed class-balancing weights for training, and BC-MultiSet.xlsx contains the patient-level clinical and molecular annotations described below. Clinical and Molecular Data Variable Description Values / Range ID Patient identifier – ER Estrogen receptor Score (0–8); Binary (0, 1) PR Progesterone receptor Score (0–8); Binary (0, 1) HER2 Human epidermal growth factor receptor 2 Score (0–3+); Binary (0, 1) MOLT Molecular subtype LumA / LumB± / HER2 / TNBC; Binary (0 - Luminal vs 1 - Non-Luminal) Nod Lymph node involvement Binary (0, 1) Ki67 Proliferation index as a percentage (%) TILs Tumor-infiltrating lymphocyte density as a percentage (%) Grade Tumor grade Score (1–3) Age Patient age years