Deep-learning-assisted biophysical imaging cytometry at massive throughput delineates cell population heterogeneity

Area of Science:

• Biophysics
• Cell Biology
• Optical Imaging

Background:

• Intrinsic cellular optical and biophysical properties are linked to health and disease.
• Label-free methods avoid costly, perturbing protocols but lack throughput and sensitivity.
• Current technologies limit the wide applicability of label-free cell assays.

Purpose of the Study:

• To develop a high-throughput platform for label-free analysis of single-cell biophysical properties.
• To enable hierarchical analysis of intrinsic morphological descriptors and mass density.
• To correlate label-free data with fluorescently labeled biochemical markers.

Main Methods:

• Developed a large-scale, integrative imaging flow cytometry platform.
• Utilized optofluidic cytometry for synchronous single-cell acquisition.
• Applied deep neural networks and transfer learning for data analysis.

Main Results:

• Demonstrated label-free delineation of cancer subtype biophysical signatures.
• Successfully detected rare cell populations (10^-5) in heterogeneous samples.
• Assessed the efficacy of targeted therapeutics using biophysical phenotypes.

Conclusions:

• The developed platform offers powerful label-free, high-throughput single-cell profiling.
• This technique can stratify physiological and pathological processes based on biophysical phenotypes.
• The approach has the potential to advance optofluidic imaging cell-based assays.