EVICAN-a balanced dataset for algorithm development in cell and nucleus segmentation.

Area of Science:

• Cell Biology
• Computer Vision
• Bioimage Analysis

Background:

• Deep learning for quantitative image analysis requires extensive annotated data.
• Current datasets often lack sufficient heterogeneity and scale for robust model training.

Purpose of the Study:

• To introduce the EVICAN dataset, a novel, partially annotated collection of cell images.
• To provide a heterogeneous training resource for deep learning in cell biology.

Main Methods:

• Curated a dataset of 4600 grayscale images from 30 cell lines.
• Included images from diverse microscopes and contrast mechanisms.
• Utilized a Mask R-CNN implementation for automated cell and nuclei segmentation.

Main Results:

• The EVICAN dataset contains approximately 26,000 segmented cells.
• Demonstrated automated segmentation with a mean average precision of 61.6% at a Jaccard Index > 0.5.
• The dataset provides unparalleled heterogeneity for deep learning model development.

Conclusions:

• The EVICAN dataset addresses the need for large-scale, heterogeneous annotated data in cell image analysis.
• This resource facilitates the development of more accurate and deployable deep learning models for biological applications.
• Automated segmentation using deep learning shows promising results on this diverse dataset.