High accuracy label-free classification of kinetic cell states from holographic cytometry

Abstract

AbstractDigital holographic microscopy permits live and label-free visualization of adherent cells. Here we report the application of this approach for high accuracy kinetic quantitative cytometry. We identify twenty-six label-free optical and morphological features that are biologically independent. When used as a basis for machine learning, these features allow blind single cell classification with up to 95% accuracy. We present methods to control for inherent holographic noise, thereby establishing a set of reliable quantitative features. Together, these contributions permit continuous digital holographic cytometry for three or more days. Applying our approach to human melanoma cells treated with a panel of cancer therapeutics, we can track the response of each cell, simultaneously classifying multiple behaviors such as cell cycle length, motility, apoptosis, senescence, and heterogeneity of response to each therapeutic. Importantly, we demonstrate relationships between these phenotypes over time. This work thus provides an experimental and computational roadmap for low cost live-cell imaging and kinetic classification of heterogeneous adherent cell populations.