SpheroidJ: An Open-Source Set of Tools for Spheroid Segmentation

Area of Science:

• 3D cell culture and cancer research
• Bioimage analysis and computational pathology
• Machine learning for biomedical imaging

Background:

• Spheroids are crucial 3D models for studying tumor behavior and testing treatments.
• Automated spheroid segmentation tools are needed but often fail to generalize across experimental conditions.
• Existing methods lack robustness when applied to diverse spheroid imaging data.

Purpose of the Study:

• To develop a versatile set of tools for automated spheroid segmentation.
• To create a robust deep learning model that generalizes across various experimental settings.
• To enhance the reliability and comparability of spheroid image analysis.

Main Methods:

• Developed a generic segmentation algorithm adaptable to different scenarios.
• Utilized the generic algorithm to reduce annotation burden for deep learning model training.
• Trained and evaluated multiple deep learning architectures, including HRNet-Seg, on diverse spheroid image datasets.

Main Results:

• The generic algorithm showed limitations in generalizing to different conditions.
• The HRNet-Seg deep learning model demonstrated strong generalization across diverse experimental conditions.
• Introduced SpheroidJ, an open-source software package for spheroid segmentation.

Conclusions:

• Developed a robust deep learning model for spheroid segmentation applicable to various imaging conditions.
• SpheroidJ provides reliable and comparable analysis of spheroids, advancing tumor behavior research.
• The tools facilitate the study of micro-environmental effects and preclinical/clinical treatments.