UPBAS-Net: An Upsampling-Powered Boundary-Aware Segmentation Network for Fluorescent Spots in Microscopy Images

Area of Science:

• Biomedical Imaging
• Computational Biology
• Cell Biology

Background:

• Accurate fluorescent spot detection and segmentation in microscopy are crucial for analyzing subcellular signals and cellular heterogeneity.
• Traditional methods struggle with delineating individual spot boundaries, especially in dense distributions, limiting quantitative analysis.

Purpose of the Study:

• To develop a unified computational framework (UPBAS-Net) for accurate, boundary-aware instance segmentation of fluorescent spots at subpixel resolution.
• To overcome the limitations of centroid localization and pixel-wise segmentation in microscopy image analysis.

Main Methods:

• Integration of Fourier interpolation-based preprocessing with an enhanced YOLOv8 architecture featuring an additional upsampling layer.
• Development of a boundary-aware instance segmentation approach for fluorescent spots.
• Creation of a user-friendly, web-based analytical platform for automated segmentation.

Main Results:

• UPBAS-Net achieved substantial improvements in spot localization accuracy, with F1-score gains up to 8.27% compared to deepBlink.
• Demonstrated excellent scalability for simultaneous segmentation of fluorescent spots and cellular boundaries, enabling single-cell resolution spatial correlation analysis.
• Provided a freely accessible web platform for nonprogrammers to perform automated segmentation.

Conclusions:

• UPBAS-Net offers a significant advancement in fluorescent spot segmentation, improving accuracy and enabling detailed quantitative analysis.
• The framework's scalability and user-friendly platform facilitate broader application in cell biology research.
• This method enhances the interpretation of complex subcellular patterns and spatial relationships within cells.