Georgia Tech and Emory University Develop Label Free Deep UV Microscopy for Nondestructive T Cell Characterization and Subtyping

New Deep-UV microscopy uses AI to identify T cell subtypes with 90% accuracy without dyes or cell damage, revolutionizing CAR-T therapy monitoring.

By: AXL Media

Published: Apr 3, 2026, 10:58 AM EDT

Source: Information for this report was sourced from EurekAlert!

Revolutionizing Immune Monitoring Through Label Free Imaging

The development of deep ultraviolet microscopy marks a significant departure from traditional flow cytometry, which typically requires the use of exogenous stains and often leads to the destruction of the sample. By utilizing a specific wavelength of 255 nanometers, which is naturally absorbed by nucleic acids, researchers at Georgia Institute of Technology and Emory University have created a high contrast imaging system for live T cells. This innovation allows for the continuous monitoring of cell cultures over time, providing a window into the life cycle of immune cells that was previously obscured by the limitations of chemical labeling.

The Role of Neural Networks in Cellular Classification

To translate these ultraviolet images into actionable medical data, the research team employed a custom residual neural network trained on samples from five human donors. This artificial intelligence model was capable of sorting T cells into three distinct categories: activated, dead, and quiescent. According to the study published in BME Frontiers, the accuracy of this AI driven classification showed an almost perfect correlation with traditional gold standard methods, achieving an R squared value of over 0.97. This level of precision suggests that the software can reliably replace more invasive and expensive laboratory procedures.

Overcoming the Challenges of T Cell Subtyping

Distinguishing between CD4 positive helper T cells and CD8 positive cytotoxic T cells proved to be a more complex hurdle than simple viability checks. While static images provided a baseline, they lacked the morphological nuance required to separate these critical subtypes. To solve this, the scientists turned to dynamic imaging, capturing 500 frame time series at a rate of 8 hertz. By analyzing the pixel wise temporal fluctuations within the cells, the team was able to quantify intracellular activity, creating a multidimensional data set that a second neural network used to achieve 90 percent subtyping accuracy.