RIVA: An Image Dataset of Conventional Pap Smear Cytology with Multiple Independent Annotations

RIVA is a curated dataset focused on conventional Pap smear cytology, developed to support research in automated cervical cancer screening and diagnostic decision support. The dataset includes a total of 959 mini-patches (image crops) extracted from real cytological samples. Of these, 386 mini-patches were independently annotated by four expert cytologists, while the remaining 573 were annotated by a single expert. In total, the dataset contains 26,158 expert annotations, with 17,716 corresponding to the multi-expert subset and 8,442 to the single-expert subset. Across all images, 7,507 unique cells were identified within the subset annotated by all experts, and 15,949 unique cells were detected overall. Annotations include lesion and cellular findings aligned with standard cytological categories, covering Squamous Cell Carcinoma (SCC), High-Grade Squamous Intraepithelial Lesion (HSIL), Atypical Squamous Cells—Cannot Exclude HSIL (ASC-H), Low-Grade Squamous Intraepithelial Lesion (LSIL), Atypical Squamous Cells of Undetermined Significance (ASC-US), inflammatory changes (INF), endocervical cells (ENDO), and Negative for Intraepithelial Lesion or Malignancy (NILM). These detailed labels enable research on diagnostic classification, inter-observer variability, lesion progression, and the development of supervised learning models in medical imaging. This dataset was created as part of the project “Feasibility study for the development of a remote and real-time diagnosis system for fixed cervicovaginal cytological smears (PAPs) assisted by Artificial Intelligence (AI)”, funded by the Instituto de Efectividad Clínica y Sanitaria (IECS, Argentina) and the Centro de Inteligencia Artificial y Salud para América Latina y el Caribe (CLIAS), with support from the International Development Research Centre (IDRC, Canada) and the Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires (UBA). The project is a collaborative effort involving researchers and students from the Universidad de Buenos Aires, Universidad Torcuato Di Tella, and Instituto Tecnológico de Buenos Aires (ITBA), in partnership with medical professionals from Hospital Bernardino Rivadavia, Buenos Aires, Argentina. RIVA serves as a valuable resource for advancing cytopathology research, improving diagnostic accuracy, benchmarking AI-assisted screening systems, and promoting the development of clinically relevant tools in global healthcare contexts.