Related Experiment Video
Updated: Jan 4, 2026
07:59
Efficient and Consistent Generation of Retinal Pigment Epithelium/Choroid Flatmounts from Human Eyes for Histological Analysis
Published on: October 28, 2022
3.3K
Deep learning predicts function of live retinal pigment epithelium from quantitative microscopy
Nicholas J Schaub1,2, Nathan A Hotaling3, Petre Manescu4
1Materials Measurement Laboratory, Biosystems and Biomaterials Division, National Institute of Standards and Technology, Gaithersburg, Maryland, USA.
The Journal of Clinical Investigation
|November 13, 2019
Summary
A new method using quantitative bright-field absorbance microscopy (QBAM) and deep neural networks (DNNs) noninvasively predicts cell therapy function and donor identity, crucial for clinical biomanufacturing.
Area of Science:
- Biotechnology
- Regenerative Medicine
- Cell Therapy Manufacturing
Background:
- Cell therapies are advancing rapidly, necessitating noninvasive methods for quality control in clinical biomanufacturing.
- Current assays for transplant function validation can be invasive or lack efficiency.
- Reliable characterization is essential for ensuring the safety and efficacy of cell-based treatments.
Purpose of the Study:
- To develop and validate a noninvasive methodology for characterizing cell therapies.
- To predict tissue function and confirm cellular donor identity using quantitative microscopy and machine learning.
- To establish robust assays for clinical biomanufacturing of cell therapies.
Main Methods:
- Quantitative bright-field absorbance microscopy (QBAM) was employed to capture cell images.
- Deep neural networks (DNNs) were trained on QBAM images to predict functional and identity parameters.
- Traditional machine learning algorithms analyzed cell shape and texture features for prediction.
- The methodology was validated using clinical-grade induced pluripotent stem cell-derived retinal pigment epithelial cells (iPSC-RPE).
Main Results:
- DNNs accurately predicted iPSC-RPE monolayer transepithelial resistance and polarized vascular endothelial growth factor (VEGF) secretion.
- The system successfully matched iPSC-RPE monolayers to their original stem cell donors.
- Machine learning algorithms identified key cellular features correlating with tissue function and donor identity.
- Noninvasive predictions aligned with established functional metrics.
Conclusions:
- QBAM combined with machine learning offers a robust, noninvasive approach for cell therapy characterization.
- This methodology can be applied to clinical biomanufacturing for quality control and functional validation.
- The findings support the advancement of cell therapies by providing reliable, efficient assessment tools.
