CPSMI2025: A curated dataset of conventional Pap smear microscopy images for deep learning-based cervical cancer
José Ocampo-López-Escalera1, Martha Rosete-Aguilar2, Héctor Ochoa-Díaz-López1
1Departamento de Salud, El Colegio de la Frontera Sur, Carretera Panamericana y Periférico Sur S/N, Barrio de María Auxiliadora, C.P. 29290, San Cristóbal de las Casas, Chiapas, Mexico.
Abstract:
Cervical cancer remains one of the most common and deathly cancers in low-resource regions, where conventional Pap smear screening - though affordable - faces diagnostic bottlenecks due to limited specialist availability. To support the development of automated screening tools, we present CPSMI2025: a curated dataset of 2169 high-resolution Pap smear microscopy images representing nine clinically relevant cytology categories, derived from >350 manually screened slides obtained from Hospital General de Zona No 2 (IMSS) and a private pathology practice in Tuxtla Gutiérrez, Chiapas, Mexico. All slides were pre-classified by both a pathologist and cytotechnologist. Images were captured using a replicable open-source low-cost microscopy platform.


