An annotated fluorescence image dataset for training nuclear segmentation methods.

Area of Science:

• Digital Pathology
• Quantitative Microscopy
• Machine Learning in Imaging

Background:

• Automated nuclear image segmentation is crucial for quantitative analysis in digital pathology and microscopy.
• Current segmentation methods struggle with variations in nuclear morphology, staining, and cell density across different tissues.
• Existing annotated image datasets are limited in scope, hindering the development of robust machine learning models.

Purpose of the Study:

• To present a comprehensive, annotated dataset for training machine learning-based nuclear segmentation algorithms.
• To provide a dataset that accommodates variations in tissue types, preparation methods, and imaging parameters.
• To facilitate the development of advanced nuclear segmentation tools for digital pathology and quantitative microscopy.

Main Methods:

• Development of a comprehensive annotated dataset featuring tightly aggregated nuclei from multiple tissue types.
• Inclusion of samples prepared using common quantitative immunofluorescence microscopy techniques.
• Demonstration of dataset heterogeneity across parameters like magnification, modality, signal-to-noise ratio, and diagnosis.

Main Results:

• The dataset includes diverse nuclear morphologies and aggregation patterns relevant to real-world applications.
• Variability in imaging conditions and sample preparation is captured within the dataset.
• The dataset is structured with suggested training/test splits and expert annotations for model evaluation.

Conclusions:

• The presented dataset significantly enhances the availability of annotated data for training nuclear segmentation algorithms.
• This resource will enable the development and validation of more accurate and generalizable machine learning models for automated nuclear segmentation.
• Facilitates advancements in quantitative analysis for digital pathology and microscopy research.