An annotated high-content fluorescence microscopy dataset with EGFP-Galectin-3-stained cells and manually labelled...

Area of Science:

• Bioimage Analysis
• Cellular Imaging
• Machine Learning for Biology

Background:

• Accurate cell detection and outline annotation are crucial for bioimage analysis, particularly in high-throughput screening.
• Developing robust algorithms and neural networks requires high-quality, annotated datasets.

Purpose of the Study:

• To introduce Aitslab_bioimaging2, a novel dataset for training and evaluating cell segmentation algorithms.
• To provide a resource with hand-labeled cell outlines for benchmarking machine learning models.

Main Methods:

• Acquisition of 60 fluorescence microscopy images of EGFP-Galectin-3 labeled U2OS cells using a Thermo Fischer CX7 system.
• Hand-labeling of cell outlines by three annotators with high inter-annotator agreement.
• Dataset includes over 2200 annotated cell objects, pre-divided into training, development, and test sets.

Main Results:

• The Aitslab_bioimaging2 dataset contains detailed cell outline annotations.
• High inter-annotator agreement ensures label quality and reliability.
• The dataset size is sufficient for training advanced neural networks for instance and semantic segmentation.

Conclusions:

• Aitslab_bioimaging2 is a valuable resource for advancing cell detection and segmentation in bioimage analysis.
• The dataset facilitates the development and validation of machine learning models for microscopy image analysis.
• Availability of pre-divided sets and potential conversion to bounding boxes enhances its utility for various object detection tasks.