Annotated Pap cell images and smear slices for cell classification

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Machine learning aids global cervical cancer detection efforts.
• Current AI research for cervical screening is limited by scarce public image data.
• Manual analysis of Pap smears is labor-intensive and requires expert interpretation.

Purpose of the Study:

• To introduce APACC, a large-scale, annotated dataset for cervical cell classification.
• To address the data scarcity hindering AI development in cervical cancer screening.
• To provide a comprehensive resource for machine learning model training and validation.

Main Methods:

• Development of the APACC dataset, comprising 103,675 annotated cell images.
• Extraction of images from 107 whole Pap smears.
• Annotation of 21,371 sub-regions within the smears for detailed analysis.

Main Results:

• The APACC dataset contains a substantial volume of annotated cell images and smear sub-regions.
• It includes detailed location information for cell images within conventional Pap smears.
• This dataset facilitates in-depth investigation and study of cervical cells.

Conclusions:

• APACC significantly expands the available resources for AI-based cervical cancer screening research.
• The dataset's comprehensive nature supports the development of more accurate and robust machine learning models.
• APACC is poised to accelerate advancements in automated Pap smear analysis and early cancer detection.