Attention-Guided Residual U-Net with SE Connection and ASPP for Watershed-Based Cell Segmentation in Microscopy...

Area of Science:

• Biomedical imaging
• Cell biology
• Computational image analysis

Background:

• Time-lapse microscopy is vital for observing cellular dynamics, but manual analysis is infeasible.
• Automated cell segmentation is crucial for quantitative analysis of large image datasets.
• Deep learning, particularly U-Net architectures, has advanced microscopy image segmentation, yet challenges persist with touching cells and low signal-to-noise ratios.

Purpose of the Study:

• To develop a robust and accurate automated cell segmentation framework.
• To improve the integration of multi-level features for enhanced segmentation performance.
• To address limitations in segmenting touching cells and low signal-to-noise images.

Main Methods:

• Proposed RA-SE-ASPP-Net, a novel deep learning architecture incorporating Residual Blocks, Attention Mechanism, Squeeze-and-Excitation (SE) connection, and Atrous Spatial Pyramid Pooling (ASPP).
• Evaluated the framework on an induced pluripotent stem cell reprogramming dataset.
• Conducted ablation studies to validate the robustness and contribution of each component.

Main Results:

• The RA-SE-ASPP-Net architecture significantly outperformed baseline models in semantic segmentation accuracy.
• Demonstrated superior performance in segmenting challenging cases, including touching cells in low signal-to-noise images.
• Achieved the most accurate semantic segmentation results on the evaluated dataset.

Conclusions:

• The proposed RA-SE-ASPP-Net framework offers a precise and robust solution for automated cell segmentation in complex microscopy images.
• The integration of advanced deep learning modules effectively addresses feature integration challenges.
• The method provides accurate cell segmentation, enabling detailed quantitative analysis for biomedical research.