ApuShape: Human-in-the-Loop Annotation Software for Fluorescent Nuclei Segmentation

Area of Science:

• Cellular Imaging
• Bioimage Analysis
• Computational Biology

Background:

• Accurate nucleus annotation is crucial for large-scale cellular imaging studies.
• Existing software for high-throughput nucleus instance segmentation is limited.
• Manual annotation is time-consuming and labor-intensive.

Purpose of the Study:

• To introduce ApuShape, an interactive software for efficient nucleus annotation.
• To enable rapid creation of high-quality nucleus annotation datasets.
• To provide a solution for high-throughput nucleus instance segmentation.

Main Methods:

• Development of ApuShape, an interactive annotation software.
• Integration of a contour-level nucleus instance segmentation algorithm.
• Implementation of a human-in-the-loop active learning strategy.
• Inclusion of a boundary refinement algorithm.

Main Results:

• ApuShape demonstrated strong performance in nucleus shape identification and refinement.
• Achieved expert-level annotation accuracy across three large-scale datasets.
• Significantly reduced manual annotation effort compared to traditional methods.
• Produced refined nucleus annotations with highly accurate contours at scale.

Conclusions:

• ApuShape provides an efficient and accurate method for nucleus annotation in fluorescence microscopy.
• The software facilitates the creation of large-scale, single-cell resolution datasets.
• ApuShape establishes a standard operating procedure for nucleus annotation in bioimage analysis.