Abstract
An exploratory analysis is presented of the APACC (Annotated Pap cell images and smear slices for Cell Classification) dataset, a large-scale collection of expertly labeled cytological images derived from 107 conventional Pap smears. The dataset includes over 103,000 individual cell or cell group images, categorized into four diagnostic classes: healthy, unhealthy, rubbish, and bothcells. Focusing on the single-cell image subset, we examined class distribution, label consistency, and inter-patient variability. The analysis revealed a substantial class imbalance, with non-diagnostic (rubbish) samples comprising the majority of the dataset and abnormal (unhealthy, bothcells) categories being underrepresented. These characteristics highlight critical challenges for model development, including the need for robust preprocessing and rare-class handling strategies. The inclusion of spatially annotated smear slices further supports context-aware classification approaches. Overall, this exploratory study establishes a baseline understanding of the dataset’s structure and diagnostic content, providing a foundation for future development of automated classification models in cervical cytology.