Enhanced cell tracking using a GAN-based super-resolution video-to-video time-lapse microscopy generative model.

Area of Science:

• Cell Biology
• Bioimage Analysis
• Machine Learning

Background:

• Cellular dynamics like growth, division, and movement are crucial for understanding biological processes.
• Time-lapse microscopy provides essential spatiotemporal data at single-cell resolution.
• Deep learning excels at cell segmentation but struggles with cell tracking due to limited annotated data.

Purpose of the Study:

• To introduce tGAN, a generative adversarial network (GAN)-based tool for synthesizing annotated time-lapse microscopy data.
• To enhance the quality and diversity of synthetic data for improved cell tracking.
• To reduce the dependency on manually annotated datasets in bioimage analysis.

Main Methods:

• Development of tGAN, a dual-resolution generative adversarial network (GAN).
• Generation of high-quality, realistic synthetic annotated time-lapse microscopy videos.
• Evaluation of tGAN-generated data's impact on cell tracking model performance.

Main Results:

• tGAN produces synthetic annotated time-lapse videos with high temporal consistency and fine cellular details.
• The generated data significantly enhances the performance of state-of-the-art cell tracking models.
• Reduced reliance on manual annotations for training cell tracking algorithms.

Conclusions:

• tGAN effectively generates realistic annotated microscopy data, addressing a key limitation in cell tracking.
• This approach improves the generalizability and performance of deep learning-based cell tracking models.
• tGAN advances the application of deep learning in bioimage analysis for dynamic cell studies.