Label-free laser scanning imaging cytometry : from instrumentation to machine learning

Abstract

Cellular imaging has been instrumental in biological and biomedical sciences for direct studies of cell morphology and their subtle changes, that intimately link to the cellular identity and its homeostatic state in health and disease/dysfunction. However, existing imaging techniques lack the ability to acquire the cellular image information at the breadth, depth, and throughput required for comprehensive profiling of the morphological characteristics of an enormous population of cells at single-cell precision – a key holding promise to unraveling the heterogeneity and diversity of cell types, states, and functions. To address this challenge, this thesis describes a series of new developments of laser-scanning imaging flow cytometry that capture millions of high-resolution single-cell images at an ultrafast imaging line-scan rate of MHz, or equivalently an imaging throughput of ~100,000 cell/sec. More significantly, these high-throughput instruments simultaneously allow high-content cytometry by analyzing high-dimensional (>100) single-cell morphological features extracted from different optical contrasts. Such a unique combination of high imaging throughput and content favorably bridges to the exploration of data-driven understanding of cellular characteristics through machine learning. First, we describe the development of time-stretch imaging flow cytometry, combined with machine-learning (specifically multi-class support vector machine (SVM)), for multi-class (14 classes) classification of phytoplankton. Operating at high imaging throughput of >10,000 cells/sec, the system enables high-accuracy classification (> 94%) simply based on the geometrics and morphologic parameters extracted from the bright-field image contrast. We then further advance the instrumentation of time-stretch imaging flow cytometry by incorporating the capability of quantitative phase imaging (QPI) – a label-free imaging technique that quantifies optical phase profile of the cell at nanometer sensitivity and accesses to the underexploited single-cell biophysical features. In contrast to conventional QPI, our technique, coined multi-ATOM, bypasses the need for interferometry for phase quantification and allows superior QPI performance, in terms of phase stability and accuracy. Employing multi-class SVM, we quantify the taxonomic and biophysical information of phytoplankton by extracting a total of 109 label-free single-cell features, including not only geometrics, morphology, granularity and but also quantitative biophysical features (dry mass and dry mass density and the spatial phase distribution, etc.) By quantifying the biophysical properties of the phytoplankton cells, we further investigate the possibility of screening of the abundance of the sub-cellular components such as lipid in phytoplankton, which is of particular interest in the biofuel, environmental research, and pharmaceutical production. Finally, this thesis describes an application of a new ultrafast laser-scanning-based imaging technique, called Free-space Angular Chirped Enhanced delay (FACED), for high-resolution, and high-throughput imaging flow cytometry. FACED imaging shows a further advancement in imaging resolution as well as the imaging contrast that are not accessible in both time-stretch or multi-ATOM imaging, i.e. fluorescence and other nonlinear optical imaging contrasts. We demonstrate the technical development of FACED-imaging flow cytometry that delivers single-cell two-photon excitation fluorescence (TPEF) and second harmonic generation (SHG) image contrasts at an ultrafast line-scan rate of MHz. As a proof-of-concept demonstration, we show that high throughput SHG imaging flow cytometry enabled by FACED demonstrates sufficient specificity for large-scale, label-free microalgae screening.