Related Experiment Video
Updated: Dec 23, 2025
08:58
Label-Free Identification of Lymphocyte Subtypes Using Three-Dimensional Quantitative Phase Imaging and Machine Learning
Published on: November 19, 2018
13.0K
Robust classification of cell cycle phase and biological feature extraction by image-based deep learning.
Yukiko Nagao1, Mika Sakamoto2, Takumi Chinen1
1Faculty of Pharmaceutical Sciences, The University of Tokyo, Tokyo 113-0033, Japan.
Molecular Biology of the Cell
|April 23, 2020
Summary
Machine learning models can identify cell cycle phases (G1/S and G2) from cell images without specific markers. This approach extracts quantitative subcellular features, offering an unbiased method for biological discovery.
Area of Science:
- Cell biology
- Machine learning
- Image analysis
Background:
- Cellular organization changes dynamically throughout the cell cycle.
- These spatiotemporal changes may contain inherent cell cycle phase information.
- Traditional methods often rely on specific cell cycle markers.
Purpose of the Study:
- To develop a machine learning approach for cell cycle phase classification using only cell morphology.
- To identify specific subcellular features that act as reliable cell cycle phase classifiers.
- To demonstrate the utility of unbiased, data-driven image analysis in cell biology.
Main Methods:
- Convolutional neural network (CNN)-based classifiers were trained on fluorescence microscopy images.
- Images were stained for nucleus, Golgi apparatus, and microtubule cytoskeleton.
- Grad-CAM analysis was used to interpret the CNN models and identify key features.
Main Results:
- Cell images were robustly classified into G1/S and G2 phases without specific cell cycle markers.
- Grad-CAM analysis identified quantitative parameters of subcellular features as effective classifiers.
- The study validated the potential of machine learning for extracting biological insights from cell images.
Conclusions:
- Machine learning-based image analysis can accurately determine cell cycle phases from general cellular morphology.
- This approach provides an unbiased and data-driven method to extract biologically relevant features.
- The findings highlight the power of AI in uncovering cellular mechanisms.
