UFSC Multi-Cyto: Multi-Stain Cytopathology Image Dataset

The UFSC Multi-Stain Cytopathology Dataset is a large, heterogeneous resource for research in cytological image analysis. It contains 17,422 expert-annotated image patches (1200×1600 or 1600×1200 pixels) with a total of 119,874 instance-level annotations, derived from 28 whole-slide images (WSIs) acquired with an Axio Scan.Z1 microscope. The dataset combines three staining protocols: Papanicolaou (oral cavity), AgNOR (cervix), and Feulgen (cervix), representing a broad spectrum of nuclear morphologies from two anatomical regions. Annotations were generated in multiple stages by specialist cytopathologists and reviewed for quality assurance. Data are provided in both COCO and YOLO formats and are divided into training (12,502), validation (2,448), and test (2,472) sets. This dataset supports the development and evaluation of segmentation models across different cytological stains and clinical scenarios. All data collection was conducted under protocols approved by the UFSC Research Ethics Committee (23193719.5.0000.0121 and 57423616.3.0000.0121).