Automated Cell Foreground-Background Segmentation with Phase-Contrast Microscopy Images: An Alternative to Machine...

Area of Science:

• Cell biology
• Microscopy
• Image analysis

Background:

• Cell segmentation is crucial for analyzing microscopy images.
• Current methods lack full automation and struggle with basic lab microscopy.
• Phase-contrast microscopy is widely used but presents segmentation challenges.

Purpose of the Study:

• To develop an efficient and automated cell segmentation method for phase-contrast microscopy images.
• To evaluate the method's performance against manual segmentation and existing software.
• To assess the method's robustness under varying imaging conditions.

Main Methods:

• An automated cell segmentation approach utilizing morphological operations.
• Performance evaluation using manual/visual counting as ground truth (156 images).
• Comparative analysis against Trainable Weka Segmentation, Empirical Gradient Threshold, and ilastik software.
• Assessment of adaptive performance under artificial blurriness, illumination changes, and varying image sizes.
• Validation using modified U-Net models trained on ground truth and generated data (16848 images).

Main Results:

• The proposed method achieved superior segmentation accuracy (Dice coefficient: 90.07, IoU: 82.16%).
• Average relative error in cell area measurement was low (6.51%).
• The method demonstrated reliability under suboptimal imaging conditions where manual analysis is inefficient.
• Comparable segmentation accuracy was achieved when training U-Net models with either ground truth or method-generated data.

Conclusions:

• The developed method offers accurate and practical automated cell segmentation for phase-contrast microscopy.
• It provides a reliable alternative to manual segmentation, especially in difficult imaging scenarios.
• The technique shows high potential for enhancing biological image analysis workflows.