Interactive Cell Segmentation Based on Active and Semi-Supervised Learning.

Area of Science:

• Computational Biology
• Image Analysis
• Biomedical Imaging

Background:

• Automatic cell segmentation is challenging for complex, long-term time-lapse microscopy data, especially without biomarkers.
• Existing methods struggle with accuracy and efficiency in intricate biological image analysis.

Purpose of the Study:

• To develop an interactive cell segmentation method that leverages human guidance for improved accuracy and efficiency.
• To address the limitations of fully automated methods in complex biological imaging scenarios.

Main Methods:

• Classifying feature-homogeneous superpixels guided by active human interventions.
• Employing transductive Rademacher complexity to select informative superpixels for annotation.
• Utilizing an affinity graph for label propagation and a verification propagation matrix for efficient error correction.

Main Results:

• The interactive method achieves high-quality cell segmentation with minimal human intervention.
• Demonstrated significant efficiency improvements compared to alternative segmentation techniques.
• Validated on diverse cell populations, showing robust performance.

Conclusions:

• The proposed interactive cell segmentation method offers a highly efficient and accurate solution for complex biological image analysis.
• Active learning strategies combined with efficient error propagation significantly reduce the need for extensive manual annotation.
• This approach enhances the reliability of cell segmentation in challenging microscopy experiments.