A rapid segmentation method of cell boundary for developing embryos using machine learning with a personal computer.

Area of Science:

• Developmental Biology
• Computational Biology
• Bioimage Analysis

Background:

• Accurate cell segmentation is vital for studying embryonic morphogenesis.
• Existing machine learning methods like U-Net are accurate but time-consuming.
• Rapid and accurate cell segmentation tools are needed to improve research efficiency.

Purpose of the Study:

• To introduce Cell Segmentator using Machine Learning (CSML), a fast and accurate cell segmentation method.
• To demonstrate CSML's performance on personal computers for embryo cell segmentation.
• To validate CSML's accuracy and generalizability compared to existing methods.

Main Methods:

• Developed CSML, a machine learning-based cell segmentation tool utilizing a Fully Convolutional Network.
• Trained the CSML classifier using a single whole embryo image and two parameters.
• Evaluated CSML's segmentation speed and accuracy (F-value) on Xenopus ectodermal cells.

Main Results:

• CSML achieved an average segmentation time of four seconds per image on a personal computer.
• CSML demonstrated a high F-value of approximately 0.97, outperforming RACE and watershed methods.
• CSML showed comparable or superior performance and speed against other machine learning methods like U-Net.

Conclusions:

• CSML offers a rapid, accurate, and efficient solution for cell segmentation in developmental studies.
• The method requires minimal training data and parameters, enhancing its usability.
• CSML is expected to significantly improve the efficiency of cell shape and developmental studies.