Bialystok Outline Annotated Cervical Cytology Dataset

Bialystok dataset contains 272 images measuring 3500×3500 px, with 4,166 cells annotated in accordance with The Bethesda System. The images are derived from 13 whole-slide images (WSI) of routine conventional cervical smears. The WSI were acquired with 40x magnification (and a resolution of 0.195 um/px) with Pannoramic 250 scanner.
and divided into non-overlapping large tiles.

 Cells present in the tiles were annotated by cytotechnologists and evaluated by an experienced pathologist. In order to train an algorithm on clear, not confusing examples, only cells for which the annotators could assign a Bethesda label with a high degree of confidence were annotated. This approach was intended to minimise label uncertainty that could affect supervised learning.
The images are accompanied by two types of annotations:  nuclei point locations and cell outlines, both with assigned labels corresponding to its diagnostic category.
The 11 categories cover the following cells:

•  Squamous epithelial:
  • Abnormal or atypical cells: HSIL, ASCUS, ASCH and LSIL, with an additional label for koilocytes,
  •  Physiological or benign (NILM) cells: parabasal, intermediate, superficial and metaplastic,
• Endocervical cells,
• Neutrophils.
  
  

All 8012 squamous cells present in the tiles have outline annotations. Cells which had no category assigned were outlined as "ambiguous".

An initial version of the dataset is described in Conventional Cervical Cytology Image Dataset with Cell Outline Annotations. This version, with additional images containing examples of  ASC-US and ASC-H, is described in Cytoplasm and nuclei as a basis for Bethesda cell cluster classification in cervical smears. Please cite the latter if you found our work useful in your research.