NuCorpus-15M: A Large-Scale DAPI-Stained Cell Nucleus Image Dataset

Dataset Description

NuCorpus-15M is a large-scale dataset specifically created for nuclear morphology analysis in biomedical research. It consists of 15.52 million high-quality DAPI-stained cell nucleus images, representing a diverse range of human and mouse tissues, both diseased and non-diseased. The dataset spans 15 different organs or tissue types and has been meticulously curated to ensure high consistency and quality for various downstream tasks such as cell type identification, perturbation detection, and gene expression prediction.

Dataset Details

• Total Images: 15.52 million
• Species: Human (78.9%) and Mouse (21.1%)
• Tissues Covered: Breast, colon, brain, tonsil, and more.
• Data Source: Images are derived from 56 standardized high-resolution tissue sections provided by 10x Genomics' Xenium platform.
• Sub-image size: 112x112 pixels, each centered around a cell nucleus.
• File format: HDF5 for both images and metadata, facilitating efficient large-scale data handling.
• Preprocessing: All images in NuCorpus-15M have been pre-processed for grayscale conversion and Normalization.
• Data Distribution: The dataset is divided into 20 parts, and the upload process is still ongoing.

Dataset Structure

Each entry in the dataset consists of a DAPI-stained image and its corresponding metadata, stored in HDF5 format to facilitate efficient access. The images have been pre-processed to normalize fluorescence intensity and remove suboptimal images.

Citation

If you use NuCorpus-15M in your research, please cite the following paper:

Hua, Y., Li, S., & Zhang, Y. (2024). NuSPIRe: Nuclear Morphology focused Self-supervised Pretrained model for Image Representations.