SiNuS: A Comprehensive Dataset for Singular Nuclei Segmentation for HER2 Grading of Breast Cancer

Singular nuclei are those that are intact and do not overlap with neighboring nuclei. According to the American Society of Clinical Oncology and the College of American Pathologists ASCO/CAP, the human epidermal growth factor receptor 2 (HER2) grading of cancer must be performed using only singular nuclei from in situ hybridization (ISH) images, regardless of cancer type. Therefore, accurate segmentation of singular nuclei is critical for automated HER2 grading in digital pathology, performed routinely for breast cancer patients. A few methods have been proposed for singular nuclei segmentation but they rely on private datasets. Although numerous public nuclei datasets are available, but they are not suitable for singular nuclei detection. We developed a comprehensive public dataset for singular nuclei segmentation. The dataset includes original annotations on Dual-ISH 20 images, binary masks, multicolor masks and nuclei-boundary masks to support diverse analysis. The annotation was done independently by three expert pathologists to ensure clinical relevance and minimize biases. Then, we included both the inclusive and exclusive nuclei selection representing combined annotation from any expert and strong consensus annotation among all experts, respectively. The dataset consists of 39 image patches comprising 1,856 inclusive and 1,284 exclusive singular nuclei, extracted from 10 HER2-negative and 11 HER2-positive breast cancer patients. The dataset was validated through experiments using deep learning–based segmentation models. Therefore, this dataset is a clinically relevant and first publicly available resource for the development and evaluation of automated singular nuclei segmentation and HER2 grading methods in digital pathology.