PI-Net: A multi-site dataset for pixel-level segmentation and staging of pressure injuries

Our dataset comprises two distinct subsets with different acquisition routes. The first subset, MIPI (Multi-Institution Pressure Injury), consists of 406 clinical wound photographs prospectively and their wound-only cropped versions. The second subset, PIID (Pressure Injury Image Dataset), was sourced from a publicly available dataset published by Betul et al. and is freely accessible at https://github.com/FU-MedicalAI/PIID with corresponding annotations provided for this dataset in PI-Net. Each image was annotated by certified wound care specialists with pixel-level segmentation masks and stage labels. By providing detailed, expert-verified annotations across the full clinical spectrum, this dataset aims to support reproducible research in pressure injury recognition and facilitate the clinical translation of AI-assisted wound assessment.