An Effective Deep Learning Framework for Cell Segmentation in Microscopy Images.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Machine Learning

Background:

• Accurate cell segmentation is crucial for analyzing cell behavior in microscopy.
• Existing methods struggle with cell segmentation in differential inference contrast (DIC) and phase-contrast (PhC) microscopy images.
• Automated and reliable cell segmentation remains a significant challenge in biological research.

Purpose of the Study:

• To develop a deep learning framework for accurate cell instance segmentation in DIC and PhC microscopy images.
• To improve the reliability and automation of cell segmentation for downstream biological analyses.
• To achieve state-of-the-art performance in cell instance segmentation on challenging datasets.

Main Methods:

• A deep learning approach combining Mask RCNN architecture with a Shape-Aware Loss function was employed.
• The framework was trained and evaluated on the DIC-C2DH-HeLa and PhC-C2DH-U373 datasets.
• No additional post-processing steps were required for generating segmentation masks.

Main Results:

• The proposed method achieved high Intersection over Union (IOU) scores: 91.91% on the DIC-C2DH-HeLa dataset and 94.93% on the PhC-C2DH-U373 dataset.
• The approach demonstrated superior performance compared to prior works in cell segmentation.
• Accurate cell instance segmentation masks were generated directly from microscopy images.

Conclusions:

• The deep learning framework provides a robust solution for cell instance segmentation in challenging microscopy image types.
• The accurate segmentation masks facilitate reliable cell behavior analysis and cell tracking.
• This method offers a significant advancement for automated biological image analysis.