fig1a_presentation.jpg
Example_prediction.jpg

(Top) Our machine learning models take phase contrast image inputs and create individual predictions of various cell marker expression profiles. These profiles can then be merged to create composite images.

(Bottom) Examples of an input, target immunofluorescence image, and a machine learning prediction image.

Investigating Heterogeneities of Live Mesenchymal Stromal Cells using AI-based Label-Free Imaging

Regenerative medicine. Immunotherapy. Tissue repair. Mesenchymal stromal cells (MSCs) hold high promise in numerous novel clinical treatments. But! Unregulated inconsistencies in the MSC phenotype negatively impact treatment outcomes. Here, we develop a machine learning model to identify MSC extracellular marker expression and localization in phase contrast images of real time, live cell cultures. Our artificial intelligence (AI)-based method converts transmitted light microscopy images of MSCs into quantitative measurements of protein expression levels. This approach allows us to perform quantitative, non-invasive, single-cell, and multi-marker characterizations of heterogeneous live MSC culture which is critical for high-throughput drug screening and quality control in cell therapies. We hope that this technique can be developed into the next gold standard as an accurate, efficient, lower-cost, real-time, and non-invasive method of evaluating cell cultures. 

Publications:

Imboden, S., Liu, X., Lee, B.S. et al. Investigating heterogeneities of live mesenchymal stromal cells using AI-based label-free imaging. Sci Rep 11, 6728 (2021). https://doi.org/10.1038/s41598-021-85905-z