The purpose of flow cytometers is to enable the classification of cells or organisms at high throughput. Label-free optical flow cytometers not based on fluorescence are generally based on scattering. The most common of these compares the amount of forward (FS) versus side (SS) scattering. Such two-parameter information permits rudimentary classification based on size or granularity, but it misses more subtle features that can be critical in defining organism identity. Nevertheless, FS/SS flow cytometry remains popular, largely because of its simplicity and capacity for high throughput. We propose to develop a label-free computational flow cytometer that preserves much of the simplicity and high-throughput capacity of FS/SS flow cytometry, but provides significantly enhanced information. Instead of characterizing organisms based on scattering direction (as does FS/SS flow cytometry), we will characterize based on scattering patterns. We will insert a reconfigurable diffractive element in the imaging optics of a flow cytometer to route user-defined basis patterns to independent detectors. The basis patterns will be optimally matched to specific sample features. The respective weights of these basis patterns will serve as signatures to identify organisms of interest. The basis patterns themselves will be determined by machine learning algorithms. Both the device and the learning algorithms will be developed from scratch. We anticipate that our flow cytometer will be able to operate at flow rates on the order of meters per second, commensurate with state-of-the-art FS/SS flow cytometers, while providing significantly more information for improved classification capacity. While our technique should be advantageous for any label-free flow cytometry application requiring high throughput, we will test it here by demonstrating high-throughput classification of microbial communities.
