Related Experiment Video
Updated: Dec 21, 2025
11:54
Establishing Embryonic Mouse Neural Stem Cell Culture Using the Neurosphere Assay
Published on: January 11, 2011
25.3K
Early Prediction of Single-Cell Derived Sphere Formation Rate Using Convolutional Neural Network Image Analysis
Yu-Chih Chen1,2, Zhixiong Zhang1, Euisik Yoon1,3,4
1Department of Electrical Engineering and Computer Science, University of Michigan, 1301 Beal Avenue, Ann Arbor, Michigan 48109-2122, United States.
Analytical Chemistry
|May 20, 2020
Summary
Machine learning accelerates cancer stem-cell identification by predicting tumorsphere formation from early images. This AI approach significantly speeds up the traditional tumorsphere assay for cancer research.
Area of Science:
- Biomedical Engineering
- Cancer Research
- Machine Learning
Background:
- Cancer stem-like cells (CSCs) are crucial for tumor progression and metastasis.
- The tumorsphere assay identifies CSCs by their ability to form spheres in suspension culture.
- Current tumorsphere assays are time-consuming, limiting high-throughput analysis.
Purpose of the Study:
- To expedite the tumorsphere assay using machine learning and single-cell analysis.
- To develop a predictive model for tumorsphere formation based on early cellular imaging.
- To enhance the throughput of CSC identification and characterization.
Main Methods:
- Collected 1,710 single-cell events for database creation.
- Trained a convolutional neural network (CNN) model using Day 4 cell images.
- The CNN model predicts tumorsphere formation by Day 14.
Main Results:
- The predictive model accurately estimated the sphere formation rate for SUM159 breast cancer cells.
- Predicted rate (17.8%) closely matched the observed Day 14 rate (17.6%).
- Demonstrated significant acceleration of the tumorsphere assay.
Conclusions:
- Machine learning integration with single-cell analysis can substantially accelerate CSC identification.
- This synergistic approach offers a powerful tool for cancer research and other biomedical applications.
- The developed model shows feasibility for rapid, high-throughput tumorsphere assays.
